Parallel nonlinear predictive control

In this work we discuss methods of implementing nonlinear predictive control on a parallel computational platform. We focus on the ability of these methods to handle, in a systematic way, large-scale systems with nonlinearities and constraints. We propose a method based on parallel linear algebra algorithms and apply this method to energy efficient control of commercial buildings. The proposed model predictive controller (MPC) minimizes energy consumption while satisfying occupant thermal comfort by using predictive knowledge of weather and occupancy profiles. We present a numerical study examining the capabilities of state-of-the-art interior point optimization solvers to utilize parallel linear algebra. We are particularly concerned with the sparsity of large scale MPC problems and how well the linear algebra algorithms are able to take advantage of that sparsity.