Order acceptance under uncertainty: a reinforcement learning approach

摘要

In this thesis, we study Order Acceptance (OA) problems under uncertainty and their solutions using Reinforcement Learning (RL). OA is an essential business problem that has not been extensively studied in this respect. Orders with different characteristics arrive stochastically at a processing facility; based on expected total profit a decision has to be made whether or not to accept an incoming order. Unaccepted orders are lost forever. RL is a promising approach that combines the ideas of modeling uncertainty and solving the problem of incomplete information (using learning methods). The idea of modeling uncertainty is based on modeling semi-Markov decision problems (SMDP).