Analytics with Stochastic Optimization: Experimental Results of Demand Uncertainty in a Process Industry

Abstract

The key objective of the research is to report the results of testing of a two stage stochastic linear programming (SLP) model with recourse using a multi scenario, multi period, menu driven user friendly DSS in a North American steel company. The SLP model and the DSS is generic which can be applied to any process industry. It is capable of configuring multiple materials, multiple facilities, multiple activities and multiple storage areas. The DSS is developed using 4th Dimension programming language, and the SLP model was solved using the IBM CPLEX solver. The value of the SLP solution derived from the experimentation of the DSS with a real-world instance of one steel mill is 1.61%, which is equivalent to a potential benefit of US$ 24.61 million. A set of experiments were designed based on the potential joint probability scenarios, and the demand distributions expected skewness. The research reports a few interesting patterns emerged from optimization results when the volatility in demand of finished steel rises and the distribution of the demand skewness changes from left to right tail. The academic contribution of this research is two folds. Firstly, the depicting potential contribution to profit in a steel company using a SLP based DSS under probabilistic demand scenarios. Secondly, the optimization experiments confirm that the value of SLP solution increases with the increase in demand uncertainty. The research has applied implications that the practicing managers would be encouraged to look for more optimization based business solutions, and the prescriptive analytics discipline will fetch more scholarly and industry attention.