Solving equation systems associated with non-linear model predictive control

Abstract: When solving optimization problems associated with non-linear model predictive control, a linear equation system of a specific structure frequently arises. Two different software, CVXGEN and FORCES, developed to generate tailored solvers to specific optimization problems use two different methods when solving this equation system. In this paper, two different sets of software is presented which generate tailored solvers for this equation system using the two different methods. It is found that the solvers which implement the method used in FORCES are faster than the solvers which implement the method used in CVXGEN, as are the generations of the solvers using the FORCES method. A third software which generate solvers to optimization problems is QPgen, which is now not able to generate solvers for non-linear model predictive control problems. The software presented in this paper is intended to be incorporated in the QPgen software in order to make it able to generate solvers for these problems as well.