Solving distributed, non-cooperative design problems using multi-agent systems.

摘要

The solution to many real world engineering problems requires decomposition of an overall system into two or more subsystems. This decomposition typically occurs along disciplinary boundaries. It is common for each of these decomposed subsystem problems to be solved by different contractors that may be different organizations, separate groups within an organization, etc. Even if they are completely independent entities, they are bound to each other by the shared responsibility to build a single artifact. As a result, the actions and decisions of each contractor are likely to have a powerful effect on the actions and decisions of the others.; This work focuses on solving the optimization problems resulting from this decomposition under the circumstances where the individual design teams are unwilling or unable to share any information regarding their analysis process. With this restriction, the final overall designs are seldom optimal in a multiobjective sense.; The primary technology used to implement this new solution technique is multi-agent systems (MAS). The technique is shown to produce solutions that perform well when compared to two benchmarks for these problems; the Nash equilibrium solution(s) and the Pareto optimal solutions. Of the five case studies presented, the Nash solutions are equaled or bested on average for all and Pareto optimal solutions are commonly achieved.