Agent-Based Modelling has now become a relevant and recognised paradigm to design integrated models of complex systems such as socio-environmental systems. The MAELIA modelling project, that aims at assessing various water withdrawal policies, is a typical example of such complex models: it couples physical dynamics (e.g. water flow and plant growth) with agricultural activities (e.g. cropping plan decision-making) to provide a Decision-Support System about water management policies. Working on such a complex model has highlighted the limits of tools and methods currently used in modelling projects. This dissertation aims at investigating more particularly three research axes that appear necessary to improve the way we design and use agent-based models.
First, the dissertation focuses on the integration of complex and cognitive agents in agent-based models. Agent-Based Models are usually designed with very simple agents and these models are generally abstract and focused on a specific process (e.g. opinion diffusion). But it has appeared necessary to integrate, in socio-environmental system models, agents able to make complex decisions (such as cropping plan decision by farmer agents in the MAELIA model) and to reason about others in large-scale artificial societies. To this purpose, a BDI architecture coupled with a multi-criteria decision-making process has been proposed and integrated in the GAMA platform. In addition, models of agents with complex social cognitive capabilities (e.g. trust and social emotions) are presented.
The second research axis deals with the integration of different paradigm models into an agent-based model, and more specifically with the coupling of Agent-Based Models with Ordinary Differential Equation models. This coupling is illustrated with the abstract MicMac model and more recently with a model of Dengue spread investigating the causal relationship between the opening of an economic corridor in South-East Asia and the number of Dengue fever cases. These models highlight (i) the benefits that the coupling of agent-based models (generative model at the microscopic level) with equation-based models (descriptive model at the macroscopic level) can bring to modellers, but also (ii) the methodological and technical difficulties of this coupling.
Finally, the last research axis focuses on issues related to data and data management in agent-based models. Agent-based models in general and socio-environmental models in particular require a huge amount of input data; they also produce a lot of data that needs to be analysed for calibration purposes or even to support decisions. To deal with these challenges, an integrated framework combining simulator, database management system and Business Intelligence tools is presented; its global architecture, implementation and application to a case study (rice pests invasion monitoring in the Mekong delta) are also detailed.
One of the main characteristics of the research activity presented in this dissertation is that all the works have been implemented in one single agent-based platform, GAMA (developed in collaboration between the IRD and several French and Vietnamese universities), that is used in numerous training sessions every year (MAPS, MISS-ABMS, JTD).
After the description of the three previous axes, the rest of this dissertation will focus on research perspectives concerning the use of qualitative data (inquiries results, testimonies, interview...) to build, feed (at initialisation) and inform (during the simulation runtime) agent-based models and simulations.
Keywords: Agent-Based Modelling, GAMA platform, BDI, trust, emotion, ODE, coupling, data management.
