Implementation issues in quadratic model predictive control

Sparse methods appear to be advantageous around the largest size of quadratic model predictive control problem being tackled commercially. Such methods may therefore become significant in the near future and deserve further study. While singular value thresholding can sometimes offer robustness advantages over simple input weighting, it does not remove the need for the latter. The two methods should be viewed as complementary and be tailored to particular applications. Further work is needed to compare the methods in the presence of constraints. The two methods are compared on a 3-input 3-output heavy oil fractionator model