Experimental evaluation of sphere decoding for long-horizon direct model predictive control

Abstract

In the past few years, the interest in model predictive control (MPC) for power electronics has increased significantly. It proves to be a good alternative to space-vector modulation at low switching frequencies. It has been shown that an increase in the prediction horizon improves the system performance. Unfortunately, increasing the horizon leads to an exponential increase in the complexity of the optimization problem. To date, most of the work that has been done is only in simulation, ignoring real-time requirements. This paper offers a practical implementation of long-horizon direct MPC. Methods on how to efficiently implement the controller on a field programmable gate array are discussed. Sphere decoding is used to solve the optimization problem. Results show that the implementation of a 5-step horizon MPC and a sphere decoder is very efficient and effective, as it requires only 8.6μs, in roughly 80% of the cases, to execute the control law with a sampling period of 25 μs when a three-phase neutral-point clamped inverter with an RL load is examined.