PITA Fiscal Year 2013 Projects

Real-Time Optimization under Uncertainty

Lead University: Lehigh University
PI: Luis F. Zuluaga, Dept of Industrial and Systems Engineering
Co-PI(s): Robert H. Storer, Dept. of Industrial and Systems Engineering
PA Industry: Air Products and Chemicals, Inc.

In producing and delivering products to their consumers, companies like Air Products use capital-intensive assets and highly-complex processes. These processes operate in a dynamic environment and in highly-competitive, rapidly-changing markets. For these reasons, Real-Time Optimization (RTO) techniques have been developed to help companies efficiently adapt assets and processes to fluctuating inputs and market conditions in a real-time fashion. However, current RTO techniques commonly assume that the problem parameters are known. This is an unrealistic assumption in the context of Air Products’ real-time asset management decisions. In fact, underestimating uncertain factors results in misleading estimates of the savings associated with the optimal solution of the RTO problem. In the context of real-time asset management in the chemical industry, classical ways to address uncertainty in decision-making problems typically lead to problems that are either too computationally expensive to solve, or whose solution is too conservative. The objective of this proposal is to address these RTO drawbacks under uncertainty by using suitable and powerful tools from Distributionally Robust Optimization, an area that has been the focus of important developments to address decision-making problems under uncertainty. An additional project objective is to go beyond customary uncertainty analysis of parameters driving the real-time decision process (demand, inventory, spot prices) and also take into account the uncertainty arising because the actual recommendations from the RTO process can only be approximately implemented.