摘要：A combination of real-time steady-state optimization (RTO) and model predictive control (MPC) used in the process industry for maximizing economic performance can result in sub-optimal operation due to the model mismatch between the RTO and MPC components and delays involved in carrying out RTO step. These problems are alleviated by recently proposed economic nonlinear MPC (ENMPC) that merges the optimization and control layers. For an e¤ective implementation of ENMPC, it is essential that accurate information of states and parameters is available at every sampling instant. The ENMPC schemes proposed in the literature either assume that the model parameters are not time-varying or are measurable/known quantities when they change with time. In practice, variations in these model parameters are seldom measurable online. In this work, two novel ENMPC formulations are proposed to address this problem: (i) Output Feedback ENMPC integrated with state estimator that employs innovation feedback for correcting model predictions and (ii) Adaptive ENMPC integrated with simultaneous state and parameter estimator, in which parameters are estimated simultaneously with the states and are updated in the model used by ENMPC. The efficacies of the proposed approaches are demonstrated by simulating two benchmark reactor control problems. Analysis of the simulation results reveals that the proposed Adaptive ENMPC is able to achieve performance comparable to the ideal case where all states and parameter variations are perfectly known. Also, Adaptive ENMPC as well as Output Feedback ENMPC formulations result in significantly better control when compared with the conventional ENMPC formulation.
