Constraints Reduction in a Multi-model Predictive Controller Applied to a Propylene Polymerization Reactor

Industrial processes are often governed by complex nonlinear dynamics, posing significant challenges for control design. While nonlinear predictive control can effectively manage such behavior, its high computational demand limits practical implementation. An alternative approach is to approximate the nonlinear system using a set of linear models within a multi-model predictive control (mMPC) framework, thereby reducing computational complexity. However, the inclusion of constraints into all models remains computationally demanding. To address this issue, two reduced-constraint mMPC formulations are proposed: one based on the static gain matrix of individual models (mMPCsg) and another on their unforced responses (mMPCur). Application to a MIMO propylene polymerization reactor – heat exchanger system demonstrates a considerable reduction in computation time while preserving control performance and maintaining constraint violations at levels comparable to the fullconstraint mMPC.