Distributed model predictive control of integrated process systems using an adaptive community detection approach

Abstract

Integrated processes offer significant advantages in chemical manufacturing, but their complexity poses challenges for control and operation. Model predictive control (MPC) shows promise in addressing these control challenges, but its practical implementation is hindered by the need for real-time optimization. Distributed Model Predictive Control (DMPC) is a viable alternative to MPC. It divides the complex process into smaller, interconnected communities and assigns a control agent to each. These agents collaborate to optimize their respective communities and share information, leading to a more manageable and efficient control strategy. This master's report thoroughly examines the relevant literature on integrated processes, distributed control and estimation, network theory, and community detection. It highlights the superiority of distributed control and estimation over centralized control, particularly when applied to unweighted networks. Furthermore, the report extends the existing research by exploring weighted graphs, generalizing state-space models to non-control affine models, and proposing novel improvements to community detection approaches. These contributions address gaps in the current literature and enhance the applicability of DMPC in real-world scenarios.