An integrated data-driven modeling & global optimization approach for multi-period nonlinear production planning problems

Abstract

In this work, we present an integrated data-driven modeling and global optimization-based multi-period nonlinear production planning framework that is applied to a real-life refinery complex. The proposed multi-period framework significantly extends and improves previous works based on single-period planning formulation by optimally managing inventories. The framework features (i) automatic generation of nonlinear and sparse data-driven process models where yields and properties of the process models are based on input properties and compositions, (ii) estimation of model parameters using two years of real-life plant data from the Daesan Refinery in South Korea, and (iii) global optimization of the large-scale nonlinear and multi-period production planning model using commercial global solvers. Computational results for multiple case studies show that the optimal multi-period plans outperform the actual plan by 57–94% in each period.