Learning of Market Prices Based on Temporal Differences for a TAC-SCM Agent

Abstract

In this work we present a learning method based in temporal differences for learning the market prices for a supply chain manager in the TAC-SCM game scenario. The agent has two learning methods based in temporal differences. In one hand he learns the short term tendency of the market to adjust its prices proportionally to the rise or drop. He also learns a value of probability to elevate the prices proportionally to the excess of real demand.