Data Driven Robust Optimization for Supply Chain Planning Models

Robust optimization for planning of supply chains under uncertainty is regarded as an efficient and tractable method, if the availability of uncertain data is ensured. Robust Optimization works by evaluating the moments of objective and constraint functions by converting the optimization problem under uncertainty into an equivalent deterministic formulation, the accuracy of which depends on the way the moments is calculated with limited amount of data. Conventional techniques such as box/budget uncertainties work by sampling in a conservative approach, often leading to inaccuracies. In this paper, machine learning and data analytics are amalgamated with robust optimization in search of efficient solutions. A novel neuro fuzzy clustering mechanism is implemented to cluster the uncertain space such that the exact regions of uncertainty are optimally identified. Subsequently, local density based boundary point detection and Delaunay triangulation based boundary construction enable intelligent Sobol based sampling leading to the evaluation of close approximations to the true moments. The proposed technique is utilized to explore the merits of robust optimization towards addressing the uncertainty issues of product demand, machine uptime and raw material cost associated with a multiproduct, and multisite supply chain planning model, a mixed integer nonlinear programming (MINLP) formulation under GAMS framewor