| Summary: | 碩士 === 國立臺灣科技大學 === 工業管理系 === 107 === This thesis primarily discusses dynamic lot sizing problems for a single item with time-varying deterministic demand. In particular, we carefully review the literature on how special cost assumptions affect the optimal policy structure so as to develop an efficient solution pro-cedure. Our main focus is placed on the development of efficient dynamic programming (DP) procedures, especially a well-known Wagner-Whitin forward DP method for seeking an ex-act optimal solution. This Wagner-Whitin DP algorithm can solve the N-period problem in O(N2) under the concave cost assumptions. In the literature, however, some statements can be found that oppose this exact Wagner-Whitin DP algorithm, misleading the readers to approx-imate heuristic procedures. This is our initial impetus for this thesis to investigate why such misperception of the Wagner-Whitin DP method has arisen in the literature.
Through our early investigation of the literature, we shall carefully describe the develop-ment of various DP procedures. To this end, we begin with a (standard) straight-forward DP that can work under no particular cost assumption, but it works painfully slow, evaluating all the possible inventory levels at each period and also all the possible decisions at each state of inventory level. In practice, however, certain cost structure often entails in the objective func-tion to be minimized. Exploiting such posed special cost structure is often useful in narrowing down the DP search space for an optimal solution. In an extreme case, where the cost func-tion and constraints are all linear functions, the standard linear programming (LP) can apply, and an optimal solution can be found at an extreme point in the feasible solution space. Un-der the concave cost assumptions, Wagner and Whitin (1958) discovered such similar solution structure, and then developed a so-called regeneration-point DP approach, which is called the Wagner-Whitin DP algorithm above.
Furthermore, Wagner and Whitin discovered what they called “Planning Horizon Theo-rem” based on the special concave cost assumption; this makes the procedure more computationally attractive than the above-mentioned Wagner-Whitin DP that works in O(N2). In the literature, some investigators appeared to ignore the special cost assumptions; in consequence, some misleading results and statements can be found that a heuristic method can outperform the Wagner-Whitin DP method for rolling-horizon schedules, some of which are totally outside the special assumptions required for the “Planning Horizon Theorem” to hold. We shall discuss this issue in this thesis.
Finally, we also describe how to improve the (standard) straight-forward DP in order to make it as attractive as the Wagner-Whitin regeneration point approach. This is motivated by Dreyfus and Law (1977, Chapter 2), where the straight-forward DP and a regeneration-point DP work equally well in solving an equipment replacement problem, although the straight-forward DP has two state variables whereas the regeneration-pint DP has only one state variable. We compare them in some detail since this comparison is new to the dynamic lot sizing literature to the best of our knowledge.
|
|---|
