Improving Primal Decomposition for Multistage MPC Using an Extended Newton method

Multistage model predictive control is a robust MPC formulation that takes into account parametric uncertainty by constructing finite set of coupled scenarios. As the amount scenarios increase so does computational cost and real-time implementation might not be possible. Scenario decomposition has been proposed to distribute computations make possible, however, typically subproblems are coordinated using steepest descent method with slow convergence properties. In this paper primal algorithm improved use nonsmooth Newtons for continuous equations. The applied gas-lift optimization system compared standard descent.