An end-to-end scheduling digital twin for multistage batch plants considering safety

Abstract

Listen

Translate article

Modern multistage batch plants, characterized by products processed through a series of stages with competing processing units, face challenges in optimizing scheduling and considering inserted events such as breakdowns and maintenance operations simultaneously in current highly dynamic and volatile market. It necessitates robust scheduling tools for efficiency, flexibility, and safety. Previous mathematical programming and meta-heuristic methods, while effective, are time-consuming for large or complex problems. Hybrid models and machine learning approaches have shown promise but often lack robustness against real-world disruptions. In response to these challenges, this work proposes an end-to-end digital twin (DT) solution based on a deep reinforcement learning model for multiproduct multistage batch plants. This model provides a versatile solution for scheduling processes of varying sizes without the need for extensive retraining, seamlessly incorporating dynamic factors. Tests on the model demonstrate its rapid interaction and robust solution capabilities, highlighting its comprehensive advantages in processing speed and result quality.