Abstract
The supply chain network optimization is a difficult problem to solve in the context of distributed (information across different members) and dynamic (changes in the structure and content of the information) environment with multidisciplinary decisions. In this paper, we address this problem from a dynamic optimization of local decisions point of view, to ensure a global optimum for the supply chain performance. This is done under the frameworks of Collective Intelligence (COIN) theory and Multi-Agent Systems (MAS). By COIN, we mean a large MAS where there is no centralized control and communication, but also, where there is a global task to complete: the global supply chain network optimization. The proposed model focuses on the interactions at local and global levels between agents in order to improve the overall supply chain business process behavior. Besides, collective learning consists of adapting the local behavior of each agent (micro-learning) to the optimization of the behavior globally (macro-learning). Reinforcement learning algorithms are used at the local level, while generalization of the Q-neural algorithm is proposed to optimize the global behavior. The model is implemented within the multiagent framework for supply chain modeling and optimization over JADE agent platform. Experimental results are discussed.