In this paper, we describe the multiagent supply chain simulation framework MACSIMA that allows the design of large-scale supply network topologies consisting of a multitude of autonomous agents. MACSIMA provides all ...
详细信息
ISBN:
(纸本)9783642041426
In this paper, we describe the multiagent supply chain simulation framework MACSIMA that allows the design of large-scale supply network topologies consisting of a multitude of autonomous agents. MACSIMA provides all agents with an adaptive negotiation module providing the fine-tuning of learning capabilities on the basis of genetic algorithms as well as of settings controlling the exchange of information about finished negotiations with other cooperating agents. On this basis the co-evolution and adaptation of price negotiation strategies as well as coalition formation processes of self-interested business agents in B2B-networked domains can be fine-tuned, simulated and evaluated. Our evaluation of the effects of self-adaptation on the overall turnover and profit of a live-tier supply network scenario shows that coordination outcome and efficiency vary significantly in dependence on (a) the elaborateness of used learn parameterizations, (b) on the homogeneity of the distribution of learn parameter settings within the agent society and (c) on the information exchange settings. System outcomes are measured on a macro-level in overall turnover, profit, and communication efficiency as well as on a group-level in the distributed group portions on turnover and profit. Our analysis shows, that an expert learn and information exchange parameterization of the agents results in an increase of the overall system outcome in sales and profit by approx. 500 percent and thus has to be fine-tuned for reaching an efficient and effective coordination outcome. Moreover, expert parameterizations result in an improved communication efficiency and outcome stability on the macro- and the group-level. Providing a subgroup of agents with superior learn capabilities results in a shift of sales and profit to the smarter agents.
暂无评论