Decomposition and Coordination for the Control of Process Networks

The optimal control of large-scale interconnected chemical plants, i.e., process networks, is a fundamental problem in the theory and practice of contemporary process control. This demands a structured and scalable approach with effective methods of decomposing process networks into constituent subsystems and coordinating the controllers’ decisions for subsystems. With the perspective of network science, community detection is proposed to reveal the underlying block structures in the network representations of control systems, which generates decompositions with reduced computational cost and retained control performance. For subsystem controller coordination, distributed optimization is adopted to guarantee the optimal solution for the monolithic system, and efficient and real-time implementable algorithms are developed to overcome the computational challenge of applying distributed optimization in control. The methods are applied to multiple benchmark processes, and a future direction in scaling up model-free data-driven control is discussed.