STOCHASTIC DYNAMIC LOT- SIZING PROBLEM WITH TOTAL QUANTITY DISCOUNT

Dynamic lot sizing problem is one of the significant problems in industrial units and has beenconsidered by many researchers. Considering the quantity discount in purchasing cost is one of the important and practical aspects of inventory problems in the field of inventory control and management, and it has been less focused in terms of stochastic version of dynamic lot-sizing problem in inventory management. In this paper, the stochastic dynamic lot-sizing problem with the assumption of existence of all-units quantity discount in purchasing is defined and formulated. Two approaches are presented to handle the solving procedure of this problem. Since the considered model is a mixed integer non-linear programing model, and the objective function of the model is the only non-linear part of the model. At first, we introduce a piecewise linear approximation model to convert the objective function to a linear term. The main solution approach breaks down the problem into four levels. At the first level, a branch and bound algorithm branches the problem on the periods with a predetermined discount level. In this case, the problem is converted to constrained version of the Sox problem [10]. This sub problem raised in each node in the branch and bound algorithm, is a mixed integer non-linear programing too, which is solved based on a dynamic programming approach in the second level. In each stage, in this dynamic programming algorithm, there is a sub -problem which is solved via a branch and bound algorithm. The problem raised in each node of the recent branch and bound algorithm is solved with lagrangian relaxation method. The numeric results found in this study indicate that the proposed approach solves the problem faster than the mathematicalprogramming model using the commercial software GAMS. Moreover, the proposed algorithm for the two discount levels is also compared with the approximate solution in the mentioned software. The results indicate that our algorithm up to 14 periods can not only obtain the exact solution, but also consume less time in contrast to the approximate model.