Optimizing model predictive control of processes for wide ranges of operating conditions

摘要

This thesis develops robustly feasible model predictive controllers (RFMPC) for nonlinear network systems and soft switching mechanism between RFMPCs is proposed to achieve softly switched RFMPC (SSRFMPC).ududThe safety zones based technique is utilized to design RFMPC by two different mechanisms i.e. iterated safety zones or explicit safety zones. Although the former one is calculated online by the relaxation algorithm and its RFMPC achieve robust feasibility, the recursive robust feasibility is not guaranteed. In contrast to the former, the latter one is calculated off-line and its RFMPC achieves recursive robust feasibility. In addition to this, the robustly feasible invariant sets in the state space are calculated off-line and the initial states need to stay inside those invariant sets in order to achieve feasible control operation.ududThe computation of RFMPC is very demanding and computing time is reduced by several methods. First, the more efficient optimization solver which is gradient type solver is used to solve the optimization task. The method to provide suitable gradients of objective function and derivatives of constraints to the optimization solver is presented. The robust output prediction is approximated and its horizon is also shortened. The optimization task is formulated in the reduced space of decision variables which is used in the implementation.ududThe proposed methodology is verified by applying to a simulated drinking water distribution systems example. Comparative simulation results are presented and discussed.ud