Generative Approaches for the Synthesis of Process Structures

In process synthesis, generative approaches are algorithmic strategies able to produce new structures, which differs from conventional optimization techniques consisting in choosing among a predetermined set of structures (e.g. heuristics and superstructure optimization). The development of these approaches has only intensified recently with the rise of both evolutionary computation and machine learning techniques. This paper aims at introducing some recent experiments, categorized into reward-driven and data-driven algorithms; and discussing key aspects of the generative steps such as: required initial database, process data representation, generative model architecture, reward design, optimization strategy and post-processing for the engineer.