Fast algorithms for constrained generalised predictive control with on‐line optimisation

This paper proposes fast algorithms for constrained generalised predictive control based on two first‐order methods, namely accelerated dual gradient projection method and fast alternating minimisation algorithm. Theoretical bounds on the number of iterations, which play an important role in the context of real‐time model predictive controllers, are provided for both algorithms. Also, some implementation issues in parallel architectures are discussed. The methods are firstly validated by simulation and their results are compared with the ones presented by commercial solvers. A three‐phase grid‐connected LCL‐filtered inverter was used as a case study. The algorithms were evaluated in an FPGA with the quadratic program computed in microseconds.