| Abstract: |
In order to assist the design and execution of multi-model experiments, we are developing a Multi-Agent System that enables researchers to abstract from the technical details of running complex model experiments and to design their studies by combining various models with ease, in an automated way. This work is done in the context of the “Post-fossil Cities” project of the Swiss National Research Programme “Sustainable Economy”. The project aims at developing a serious game that will enable stakeholders to simulate possible pathways to reach carbon neutrality in a fictional Swiss city. The game is required to address both natural environment and societal constraints, while enabling stakeholders who are playing the game to identify and test different trajectories from the present state to carbon neutrality. To reach the project goals, the project itself is divided in three major parts (i) the serious simulation game, (ii) the socio-economic metabolism model, which is the core component of the software supporting the game and will represent the dynamics of the main material stocks and flows of Switzerland and (iii) the software system linking the role-playing (simulation) game with the background computer (simulation) model. This last component integrates the other two, and is the research topic of the presented work. Decisions taken by the players will be transformed into model experiments which will determine the consequences of the entirety of the decisions of all players in future years. The core model for the automated experiments is the socio-economic metabolism model mentioned above, constructed using the Material Flow Analysis (MFA) methodology. Additional models will be used to provide further details that are not directly available in the core model. One of the main challenges of this project is the overwhelming complexity of the task. The heterogeneity of the models necessary to simulate a country on its path to a post-fossil state makes this project an interesting software engineering problem. This leads to a first research question (RQ), RQ1: What strategy is successful in integrating models of different types with a role-playing game? The second challenge relates to the game design. In specific, the game is planned to evolve and must be flexibly configurable to accommodate different “families of games”. That is, even some of the basic mechanics, such as the handling of time during experiment execution, must be configurable. This flexibility comes at the expense of a large design space for the software system, and thus leads to RQ2: What software architecture is flexible enough to accommodate the large design space of such a modelling project? A third challenge of this project relates to a methodological aspect of simulation. The conceptual architecture of the role-playing game means that users, i.e. players, of the software system will be conducting model experiments, albeit probably not totally aware of the fact. These model experiments have to be defined by game designers and appear in the game as “action cards”. However, it is simply impractical to manually define all of the potential questions that appear in the game, specially when placed into context of an existing model state. This brings about RQ3: How to provide a higher abstraction level to specify complex experiments with simulation models? These three research questions led up to a preliminary software architecture, using the Multi-Agent Systems paradigm. The idea is to provide software defined agents that take care of the laborious model experiments tasks, including handling coordination of different models, and allow experiments to be defined in a higher abstraction language. The system is composed of different agents that handle different tasks. For example, domain-specific agents which handle specific model types, to coordinator agents that handle model coupling, or verifier agents who ensure consistency and verify constraints. This approach, defined here as “Agents Using Models”, brings about a flexible system, containing encapsulated and pluggable entities, which leverages ubiquitous computing power in order to assist researchers in experimentation with models. |