Study on Distributed Decision Making by Means of a Market Oriented Model

点击翻译按钮获取中文摘要

This research program has investigated 'the market oriented model' that carries out resource assignment among software agents for fast and large scale decision making. Methods for the model and their effectiveness has been examined taking a problem of assigning communication band width as an example.First, we have formulated the market oriented model for communication band width assignment. In the formulated model, communication demands have been formulated as two sorts of demands. One is a demand type that is real-time but elastic. The other is less elastic but allows substitution among time-of-use. To make software agents manage these demands, a forward market model that lists band widths in several time slots to a prescribed future is constructed. Based on this formulation, we have carried out implementation of the model into computer code and evaluation of the model through computer simulation. As a core part of the market model, we have compared Walrasian type model, fixed point search algorithm for a multiple goods market, and call auction used in a real security market. Computer simulation shows that the first two methods find equilibrium prices accurately but are expensive in computation costs. Contrary to this, the last one is practical in computational load, but is less accurate and has difficulty in equity among the agents.Further, we have also investigated interaction among agents in the market consisting of relatively small number of agents. On this issue, we have carried out analysis of the fundamental properties of such market combining oligopolistic market model and machine learning model.