Fast Generalized Predictive Control Based on Accelerated Dual Gradient Projection Method

Abstract

In general, model predictive control (MPC) requires the computation of a quadratic programming problem (QP) at each sampling instant. This computation can be considered costly from the computational point of view and become a limitation for the use of MPC in plants with fast sampling rates. In order to circumvent this limitation and allow it to be used on a larger variety of systems, special solvers which efficiently compute the control signal can be used and implemented using high-speed hardware. Several works were proposed for this type of solution, but most of them focus on state-space formulations for MPC, which are very popular in academia. This paper proposes a solution based on the accelerated dual gradient projection (GPAD) method, applied to Generalized Predictive Control (GPC), which is a very popular formulation in industry. The method is firstly validated using MATLAB and its results are compared with the ones presented by quadprog solver. A small size system is also evaluated in an FPGA with the QP computed in microseconds.