Integrated deep learning-production planning-economic model predictive control framework for large-scale processes. A fluid catalytic cracker-fractionator case study.

In this paper, we seek a tighter coordination between the production planning and process control layers in the process decision-making hierarchy, with the purpose of reducing the economic gap between planned and realized production. We introduce deep learning-based frameworks that support a stochastic production planning (PP) formulation that integrates economic model predictive control (EMPC) and PP via a feedback mechanism that relays the effect of process disturbances (traditionally dealt with by the control layer) to the planning layer. Our framework demonstrates superior economic performance, promising solution times (potentially allowing industrial implementation) and a close match of planned and realized process economics compared to a traditional industrial linear and non-integrated EMPC-PP benchmark. • A deep learning/hybrid EMPC structure is presented. • An interpretable, integrated EMPC and production planning framework is developed. • Framework is demonstrated on a representative large-scale process. [ABSTRACT FROM AUTHOR]