RUTCOR Seminar - September 27, 2007

 

Christian Almeder

University of Vienna Bruenner Str. 72, A-1210 Vienna, Austria

 

Combining ant colony optimization and exact methods for multi-level capacitated lot-sizing problems

 

Abstract: We developed 2 new solution methods for multi-level capacitated lot-sizing (MLCLS) problems. This method is a combination of a MAX-MIN ant system and an exact LP/MIP solver. The MLCLS is a part of the material requirements planning (MRP) process and it describes the problem of determining a production schedule for a set of products including the preliminary products and raw materials. That means to calculate the amounts of intermediate items and final products that should be produced in each period. There are several constraints and objectives that should be considered while developing a production schedule:

 

(i) The given demand of the final products has to be fulfilled, no backorders or shortages are allowed, but unused products are stored for later use;

 

(ii) each production decision causes resource needs for setting up production (independent of the amount produced) and for the production of each item;

 

(iii) each production step needs a deterministic amount of time (measured in periods);

 

(iv) costs occur for setup, production and holding.

 

This kind of problem is NP-hard and usually real world applications are too large to be solved

exactly. But small mixed-integer problems or pure linear problem without any binary decision

variables can be solved very efficiently.

 

For the first approach we use the ant system to optimize the decomposition of the problem into smaller subproblems. These subproblems containing only a few items and periods are modelled as mixed-integer programs and solved exactly using CPLEX. Then the overall solution is derived by consolidating the partial solutions. This hybrid approach provides superior results with respect to solution quality in comparison to the existing approaches in the literature.

 

For the second approach the solution method is split into two levels. On the first level we use a MAX-MIN ant system, for selecting the production periods for each item. Each ant starts with the end product and determines for each period, if there should be production or not, but the amounts to be produced and the amounts to be stored are still open. The calculation of these decision variables is done on the second level. This second level problem is a rather simple linear problem without any binary or integer decision variables. We use CPLEX for solving this problem exactly. To improve the solution further, a local search method is applied in two steps.

 

We tested both methods using available test instances and compared the results with those of other methods published in literature. It turned out that the first approach provides superior results with respect to solution quality in comparison to the existing approaches. But for medium-sized instances the second approach delivers even better results.

 

C. Almeder. A hybrid solution approach for multi-level capacitated lot-sizing problems. Working paper, 2007.

 

R. Pitakaso, C. Almeder, K.F. Doerner, and R.F. Hartl. Combining population-based and exact methods for multi-level capacitated lot-sizing problems. International Journal of Production Research, 44(22):4755–4771, 2006