A multi-agent system of adaptive production networks

Production networks of firms linked by supply-customer relationships embedded in a geographical space are among the phenomena not yet well understood by practitioners and scientists. A production network can be defined as a network of autonomous or semiautonomous business entities collectively responsible for procurement, manufacturing and distribution activities associated with one or more families of related products. Such networks are highly non-linear and exhibit complex behavior through the interplay of their structure, environment and function, this complexity making it difficult to manage, control or even predict them. Production networks, supply chains and their management have received considerable attention from researchers in various disciplines over the past two decades. Agent-based modeling (ABM) and simulation are regarded as one of the best candidates for addressing different aspects of these networks. Indeed, ABMs allow the study of complex systems such as production networks, from a micro-macro evolutionary modeling perspective. They are able to take into account both the issue of heterogeneity and autonomy of the agents, the relevance of their temporal-spatial dynamic relations and the emergent evolutionary nature of collective phenomena. In this paper, we present a Multiagent system for studying production networks dynamic. we propose a MAS starting with the model proposed byWeisbuch and Battiston [3], and try to understand how the behavioral styles at micro-level (agent-level) determine the proximity relations at macrolevel. In W&B’s Model, some strong assumptions regarding the regularity of the network, the orders from the market, the non-existence of pricing mechanisms and investment strategies, make the model very far from reality. Indeed, the firms are considered behaving uniformly, while in real market, firms are very heterogeneous, behave in different ways and scale of time and space. Our approach in this work is to start with this simple model characterizing well known stylized facts and use MAS techniques and paradigms in order to have a more realistic and flexible model allowing us