This thesis is to develop rolling horizon procedures for the following four cases of dynamic lot-size models.
First, we consider a production planning model for a single-product single-facility problem with two complicating factors of the production and inventory cost. The problem is analyzed under the assumption that holding and setup costs are time-variant. The problem also allows that the marginal cost for production above a specified quantity is less than that for production below the quantity.
Second, we consider a production planning model for a multi-product single-facility problem where backlogging is not allowed, and a single input resource is employed. Furthermore, in each production period the facility (or plant) produces many products simultaneously, each representing a fixed part of the involved production activity (or input resource quantity).
Third, we consider a production planning model for a single-product single-facility problem where capacity restrictions are imposed on production, and backlogging is not allowed.
Finally, we extend the second model by limiting available capacity in each period.
In each case, the optimal solution properties are derived and then used in developing a forward dynamic programming algorithm for solving the problem with finite horizon.
In the first and second cases, the corresponding planning horizon theorems which guarantee that the first-period decision will not change no matter how much information is subsequently added are proved. In case planning horizon is not obtained, a systematic procedure for selecting the first-period decision is proposed. A set of simulation experiments is then performed to investigate the cost effectiveness of the proposed procedure in rolling horizon environment. The computational results demonstrate that the proposed procedures are cost-effective.
In both the third and fourth cases, two rolling horizon procedures for first-period decision are proposed. One is based on the optimal s...