| Summary: | This paper introduces a novel fast model predictive control (MPC) methodology based on linear parameter-varying (LPV) systems. The proposed approach can deal with large-scale problems better than conventional fast MPC methods. First, the equality constraints given by the model equations are not eliminated to get a condensed quadratic programming (QP) problem, as the model of the LPV system changes and it will be time-consuming to reformulate the QP problem at each sampling time. Instead, the proposed approach constructs a sparse QP problem by keeping the equality constraints. Although the resulting QP problem has a larger dimension than the condensed one, it can be reformulated and solved as a system of piecewise affine equations given by the Karush-Kuhn-Tucker conditions of optimality. Finally, the problem will be solved through a Newton-method and an exact line search in a fast way. The performance is tested and compared with off-the-shelf QP solvers on the conventional buck dc-dc converter control problem both in simulations and the experiments on FPGA. The proposed methodology works well for the controller and is especially faster in comparison with some other conventional algorithms for large prediction horizons.
|
|---|
