Designing the process designer: Hierarchical reinforcement learning for optimisation-based process design

Optimisation-based design is an established methodology that aims to achieve a globally optimal solution complex process task by representing it as optimisation problem. We propose hybrid framework for decomposition-based design, centred around hierarchical reinforcement learning and mathematical programming. The enables the agent assemble processes, employ programming, discover designs without need pre-defined superstructure. composed of: (i) higher level, learns construct overall connecting sections, (ii) lower build solve sections initialising unit operations. Such modularity allows flexible robust in constrained, nonlinear nonconvex spaces. demonstrated case study of intensified ethylene oxide production plant, yielding improved results compared baseline reported open literature. was implemented Pyomo. Results reveal insights on agent’s speed ability leverage models.