Abstract

This article explores the design and analysis of deep neural network-based approximators for constrained finite control set model predictive control (MPC) with long prediction horizons in power electronics applications. The study focuses on the impact of the prediction horizon on the penalization of switching effort and examines how these approximators inherit operational characteristics from traditional finite control set MPC controllers. The integration of deep neural networks aims to reduce the computational demand typically associated with finite control set MPC, while ensuring high fidelity in approximating the optimal control policy. The performance is evaluated through metrics such as total harmonic distortion, switching effort, voltage quality using fast Fourier transform analysis, and execution time. Simulation scenarios include both linear and nonlinear loads in uninterruptible power supply systems, highlighting the effectiveness and adaptability of the proposed approach in modern power electronics applications.