A machine learning-based methodology for multi-parametric solution of chemical processes operation optimization under uncertainty

Chemical process operation optimization aims at obtaining the optimal operating set-points by real-time solution of an problem that embeds a steady-state model process. This task is challenged unavoidable Uncertain Parameters (UPs) variations. MultiParametric Programming (MPP) approach for solving this challenge, where must be updated online, reacting to sudden changes in UPs. MPP provides algebraic functions describing as function UPs, which allows alleviating large computational cost required each time UPs values vary. However, applicability requires well-constructed mathematical process, not suited optimization, complex, highly nonlinear and/or black-box models are usually used. To tackle issue, paper proposes machine learning-based methodology multiparametric continuous problems. The relies on offline development data-driven accurately approximate behavior over space. developed using data generated running original complex under different values. are, then, used online to, quickly, predict solutions response variation. applied benchmark examples and two case studies optimization. results demonstrate effectiveness terms high prediction accuracy (less than 1% NRMSE, most cases), robustness deal with problems natures (linear, bilinear, quadratic, black boxes) significant reduction complexity procedure compared traditional approaches (a minimum 67% time).