Agent-Based Spatial Simulation with NetLogo Volume 1

Arnaud Banos, Christophe Lang, Nicolas Marilleau

• Publisher: ISTE Press - Elsevier
• Year: 2015
• Print ISBN: 9781785480553
• eBook ISBN: 9780081007235
• Topic: Computer Science
• Subtopic: Artificial Intelligence (AI) & Semantics
• Index: Computer Science

Agent-based modeling is a flexible and intuitive approach that is close to both data and theories, which gives it a special position in the majority of scientific communities. Agent models are as much tools of understanding, exploration and adaptation as they are media for interdisciplinary exchange. It is in this kind of framework that this book is situated, beginning with agent-based modeling of spatialized phenomena with a methodological and practical orientation. Through a governing example, taking inspiration from a real problem in epidemiology, this book proposes, with pedagogy and economy, a guide to good practices of agent modeling. The reader will thus be able to understand and put the modeling into practice and acquire a certain amount of autonomy. - Featuring the following well-known techniques and tools: Modeling, such as UML, Simulation, such as the NetLogo platform, Exploration methods, Adaptation using participative simulation

1

Introduction to the Agent Approach

Fabrice Bouquet; Sébastien Chipeaux; Christophe Lang; Nicolas Marilleau; Jean-Marc Nicod; Patrick Taillandier

Abstract

When we need to study a real system made up of interconnected elements, where each of these systems has its own dynamics, it is often impossible to foresee the emergence of a global dynamics for the system. In this case, what is in question is a complex system, because any one modification, even if it is marginal in terms of its one or several constituent elements, may lead to a dramatic change in overall operation of the system. It becomes clear that these phenomena may well be understood and observed only through the construction of a model. Even if in certain particular cases the model may be resolved analytically, as is the case for the Lotka–Voltera prey-predator models, computer simulation is indispensable in all other cases, i.e. in most thematically interesting cases. As such, agent modeling is one possible response for studying complex spatial systems.

Keywords

Agent Communication Language (ACL)

Agent group role (AGR) model

Belief desire intention (BDI)

Gravity models

Multi-agent systems (MAS)

Organization-centered multi-agent systems approach (OCMAS)

Simulation models

Spatial modeling

Statistical and econometric models

1.1 Introduction

When we need to study a real system made up of interconnected elements, where each of these systems has its own dynamics, it is often impossible to foresee the emergence of a global dynamics for the system. In this case, what is in question is a complex system, because any one modification, even if it is marginal in terms of its one or several constituent elements, may lead to a dramatic change in overall operation of the system. It becomes clear that these phenomena may well be understood and observed only through the construction of a model. Even if in certain particular cases the model may be resolved analytically, as is the case for the Lotka–Voltera prey-predator models [VIA 11], computer simulation is indispensable in all other cases, i.e. in most thematically interesting cases. As such, agent modeling is one possible response for studying complex spatial systems.

Multi-agent systems (MAS) originally came into existence in the 1980s, at the crossroads of Distributed Artificial Intelligence¹ (DAI) and Artificial Life² (A-Life) [FER 95], and are currently extremely popular. What is unique about them is their capacity to make apparent collective behaviors resulting from individual actions and interactions [JEA 97].

Within the domain, MAS are viewed as an entirely simulatory approach, which complements traditional techniques based on analytical, stochastic or other types of models [VAR 13]. As with the object concept [BOO 91], MAS engage a process of structuring thought which helps researchers or those involved in industry to solve the various problems they face. MAS are considered as the logical continuation of the object concept [FER 95, WOO 97] which brings increased modularity due to its ability to adapt, to learn and to be autonomous.

The fundamental principle upon which the multi-agent paradigm is based is that of breaking down complex objects into new, smaller problems, which are easier to model [BER 05]. Thus, the agent paradigm is “more a way of thinking than an implementation technique” [FOU 05]. It simply organizes our thought by analogy with the world around us. It is an elegant and intuitive way of envisaging and representing a complex phenomenon. In fact, this is one of the reasons why this approach has been adopted in a wide range of disciplines such as social sciences, ecology and finance, among others.

In this chapter, we will introduce the concept of the multi-agent system, beginning with a presentation of two examples of the use of such systems, in social sciences and soil sciences. We will then discuss the major trends in modeling and situate agents within the context of this work. Following this, we will formally define MAS before finally applying them to two concrete examples.

1.2 Two different MAS shown through examples

In order to illustrate the extraordinary expressive capacity of MAS, we have selected two concrete case studies which lead to two diametrically opposed models. The first case study, from social sciences, is concerned with the mobility between towns of town-dwellers, and the second case study, from ecology, studies soil sustainability.

1.2.1 MAS in social sciences

The agent paradigm is a modeling approach which is very well suited to the representation of the human being as an autonomous, intelligent individual, who is capable of learning and communicating with others. The agent approach also offers the advantage of providing a natural representation of the individual [SAN 05]. As a result, the model can be used to address a research question which may apply to a range of disciplines, such as the sustainability of a town, through one of its core components, daily mobility. If we consider a town to be a form of spatial organization, it is one that provides conditions which favor social interaction. As such, it follows that an ever-increasing number of daily transport journeys are required in order to achieve the objective of linking places. Urban spread, the functional specialization of urban areas and the low social value placed on mobility are factors that contribute to this trend and intensify its effects.

The right to a given level of mobility, linked to the desire for increasingly individualized and autonomous lifestyles, may result in significantly reduced accessibility of the town and its services. Given that “too much mobility kills a town”, if a town’s development has to be harmonious and sustainable, then researchers need to identify the conditions according to which daily and individual mobility do not prevent the town from fulfilling its role. This study must also take into account the management of urban growth, which nowadays causes many problems such as urban spread, congestion, energy consumption and production, and risks and dangers to the population.

Under these conditions, it is natural and logical to use the agent approach to model individuals who move around the urban zone according to timetables and certain socioeconomic characteristics: the city-dweller is an intelligent agent attempting to carry out a series of activities; the town is an environment regulated by transport and traffic rules. This approach is quite similar to a city-dweller/agent bijection.

1.2.2 MAS in soil sciences

The versatility of the MAS means that the approach is totally malleable, and its use may be adapted at will for the case study or research question to which it is applied: this approach may even be implemented for the study of soils.

Soil is a key component of ecosystems, which is the support for one of the main ecosystem services: the production of biomass (food, fodder, energy, wood and fibers). It is a critical resource, and one that is under threat. It is also non-renewable. As a result, it is essential to promote sustainable management practices both to halt its degradation and to foster its rehabilitation. New techniques for the rehabilitation of soil, such as soil building, are currently being developed. In order to evaluate the level of ecosystem service that a rehabilitated soil can provide, and in order to predict its development and sustainability, computer modeling and simulation tools are required. However, the multi-level character of soils and the overlapping of ecosystem processes involved within them mean that it is often difficult to model their complex systems using a classical macroscopic approach. In fact, soil is characterized by both biotic processes (linked to living organisms such as earthworms) and abiotic processes (linked to non-living elements such as the physics of the materials and the flow of liquids), which interact at various levels, from macro-fauna (such as termites and earthworms) or macro-aggregate (such as silt or clays) to microbes and the micro-structure (clay) of soil. Therefore, it is necessary to use modeling approaches which can handle this overlapping of various levels.

Agent-based modeling is particularly well suited to this context. For example, the Sworm model [MAR 08] describes a dynamic three-dimensional space in the form of a fractal made up of cubic cells. Each cell assumes the role of a soil aggregate with a particular behavior. As such, each cell can be represented by an agent with its own dynamics and its set of interactions, irrespective of its size. Its behavior can be driven by submodels such as the Mior model for the decomposition of organic material [MAS 07] or other models for water retention.

In contrast to the preceding model, the agents no longer represent individuals within a space under study, but rather they represent portions of space animated by biological processes. Due to the sheer number of microbes, it is technically impossible to represent each of them by an agent. Also, the current state of knowledge on microbial individuals means that it is impossible to define behaviors at their level.

1.2.3 Summary

It can be seen through these two examples, respectively, from social and soil sciences that MAS are malleable as a function of the context, the state of knowledge of the real system and the underlying research question. The fact that they are different helps our ...

Table of contents

• Cover image
• Title page
• Table of Contents
• Copyright
• 1. Introduction to the Agent Approach
• 2. Description Formalisms in Agent Models
• 3. Introduction to NetLogo
• 4. Agent-Based Model Exploration
• 5. Dynamical Systems with NetLogo
• 6. How to Involve Stakeholders in the Modeling Process
• Bibliography
• List of Authors
• Index