Novel Feasible Set Learning and Process Flexibility Analysis Method Using Deep Neural Networks

Abstract

The operational flexibility of a chemical process refers to its ability to maintain feasible operations despite uncertain deviations from the nominal conditions. It is an important characteristic that ensures the system’s adaptability and resilience in the face of changing operating conditions. To quantify the feasible region and evaluate the flexibility of a given process design, the volumetric flexibility index is used by calculating the ratio between the hypervolume of the feasible region and the hypervolume of the region that encompasses all possible combinations of expected uncertain parameters. To deal with general problems involving nonlinear constraints, nonconvex, nonsimply connected, or high-dimensional feasible regions, we introduce a novel method that utilizes a deep regression network and a classification network to achieve a reliable and efficient evaluation of the flexibility index. We demonstrate the effectiveness of the proposed method through multiple numerical illustrations and case studies.