An SKU decomposition algorithm for the tactical planning in the FMCG industry

Abstract

In this paper we address the optimization of the tactical planning for the Fast Moving Consumer Goods (FMCG) industry, in which numerous trade-offs need to be considered over possibly thousands of Stock-Keeping Units (SKUs). An MILP model for the optimization of this tactical planning problem is proposed. This model is demonstrated for a case containing 10 SKUs, but is intractable for realistically sized problems. Therefore, a decomposition algorithm based on decomposing the model into single-SKU submodels is proposed in this paper. To account for the interaction between SKUs, slack variables are introduced into the capacity constraints. These slack variables initially allow the capacity to be violated. In an iterative procedure the cost of violating the capacity is slowly increased, and eventually a feasible solution is obtained. Even for the relatively small 10 SKU case, the required CPU time could be reduced from 4427s to 472s using the algorithm. Moreover, the algorithm was used to optimize cases of up to 1000 SKUs, whereas the full model is intractable for cases of 25 or more SKUs. The solutions obtained with the algorithm are typically within a few percent of the global optimum.Keywords: Tactical Planning, Optimization, MILP, Decomposition Algorithm, Fast Moving Consumer Goods IntroductionThe scale and complexity of enterprise-wide supply chains has increased significantly due to globalization. (Varma et al., 2007) Recently, the operation of enterprise-wide supply chains has attracted much interest. Grossmann (2005) and Varma et al. (2007) review the current research on Enterprise-wide Optimization (EWO), and they identify challenges and research opportunities. One of the main challenges is the integration of decision-making across various layers. This includes the integration of the various echelons of the supply chain and the integration of the various temporal decisions layers. The decisions on the various layers are often interconnected leading to trade-offs between these decisions. (Maravelias and Sung, 2009) Therefore, better solutions can be obtained if these decisions are optimized simultaneously.Usually, three temporal decision layers are distinguished: strategic planning, tactical planning and operational planning. Strategic planning covers the long-term decisions regarding the design of the supply chain. Tactical planning covers the medium-term decisions regarding the allocation of capacity. Operational planning covers the short-term scheduling decisions.Maravelias and Sung (2009) review the integration of short-term scheduling and tactical production planning. They identify two options for this integration. First, the detailed scheduling decisions can directly be included into the tactical planning model. While this would in theory yield optimal solutions, the resulting models are usually very large and difficult to solve.

Multistage Production Planning in the Dairy Industry: A Mixed-Integer Programming Approach

“…The model was demonstrated on an ice cream scheduling problem. In a very recent work, van Elzakker et al 29 proposed an MILP model to address optimization of the tactical planning for the FMCG industry. To solve these extremely large problems, they proposed an algorithm based on SKU decomposition.…”

A systems engineering perspective on process integration in industrial biotechnology

“…If the logistics of the distributed feed supply and products delivery to customers are tightly coupled to the manufacturing, then production planning and scheduling become intertwined with the process operations, while seasonal dynamics also play an important role. Such (desirable) integration presents a formidable optimization challenge, as shown by Van Elzakker et al for FMCG.…”

Product decomposition strategy for optimization of supply chain planning

“…Calfa et al (2013) combined temporal Lagrangean decomposition with bilevel decomposition for the integration of planning and scheduling. The opportunity of decomposing the problem by subproblems is also identified by van Elzakker et al (2014). However, they propose a heuristic procedure, while we generalize these ideas using Lagrangean decomposition.…”