A two phase MPC and its application to a grinding process

The growing complexity of the control systems and the increased use of nonlinear models cause a dramatic increase in the computational requirements of MPCs. Therefore, more computationally efficient MPC are needed. This paper presents a two-phase MPC approach for decreasing computational demand without sacrificing its efficiency. The first phase of the MPC treats the input variables as independent decision variables of the objective optimization, since the largest part of the objective value arises from a few earliest sampling intervals. In contrast, the second phase combines input variables, defining the rest of the MPC objective value, in an open-loop control which is specified by a few independent decision variables. The method is compared against the traditional Quadratic Programming implementation of an MPC for the Grinding Plant control problem. The two-phase MPC demonstrates a better performance compared with the traditional controller with the same control horizon.