A Frequency Based Transit Assignment Model That Considers Online Information ...
Abstract ENG
1. In this thesis, an approach is developed to obtain an optimal pricing policy for chartered flights. In
order to do so, a model within the framework of dynamic programming is presented and its main
structure is also analyzed. Since in real world cases the dimension of this model happens to be
very large, a solution method is developed by “Q Learning” technique. This is an appropriate
approach in approximate dynamic programming and reinforcement learning. Analysis is carried
out under two different assumptions regarding demand, namely “linear-deterministic” and
probabilistic demand for transition probabilities. An exact solution for deterministic demand case
is developed. Furthermore, for probabilistic demand also an approach within Q Learning
technique is proposed. In this approach the concept of reservation price prices is also considered.
To illustrate the proposed method, we implement it with real sales data of Tehran-Mashhad flights.
We also show the outstanding performance of Q learning for the deterministic demand function,
if the number of iterations is large enough. We show that the proposed model for the probabilistic
demand, in the presence of reservation price, is able to simulate high and low traveling seasons.
The performance of exploration versus exploitation in the mentioned environment is also
evaluated.