Abstract:
It is widely accepted that model predictive control
(MPC) with long prediction horizons yields, in general, a better
performance than with short horizons. In the context of power
electronic systems, the main advantages include improved closedloop
stability and lower current distortion per switching frequency.
A shortcoming of MPC with long prediction horizons is
the computational burden associated with solving the optimization
problem in real time, which limits the minimum possible
sampling interval. The solution to the MPC optimization problem
is a polyhedral partition of the state-space. Pre-processing of
the state-space and storing representative information thereof
offline assists in reducing the online computational burden. The
problem structure is a special case in the form of a truncated
lattice. Exploiting this characteristic enables representation of
the partitioned space to be is stored as a minimal set of Voronoi
relevant vectors describing the basic Voronoi cell of a lattice. We
evaluate the algorithm proposed by Micciancio and Voulgaris
known as the MV-algorithm to solve the closest vector problem
with pre-processing (CVPP). The performance of the algorithm is
evaluated in a simulated three-level neutral point clamped (NPC)
voltage source inverter with an RL load.
