Aggregating Demand and Product Data

Products not modeled explicitly were clustered into groups of similar products according to their raw material consumption profiles. To account for the fact that non-chemical process steps can also constitute a considerable share of the processing activities, within each of the clusters obtained this way, sales price clusters were formed if required. The assumption underlying this approach is that for products with similar raw material consumption sales prices are correlated to the complexity of non-chemical processes, whereas raw material consumption profiles are sufficient to assess the complexity of the chemical processes. 

Due to the nature of the product portfolio considered in the pilot application, it was possible to create the clusters rather intuitively without using a formalized methodology such as cluster analysis. When needed clustering techniques such as those used to form manufacturing cells could be adapted to cluster a more complex product portfolio (see Selim et al. (1998) for a review of different clustering techniques). An application of cluster analysis to aggregate demand locations is reported in Muller (1983, pp. 188-190). An intuitive approach similar to the one proposed above was also used by Nickel et al. (2005, pp. 168-170) for an application in steel industry. 

---