Abstract

Power converters are specialized systems that transform electrical energy within industrial application. Although advances in conventional control systems have driven increasingly optimal power converter operation, the efficiency of current topologies can be reduced significantly due to their intricate nature, susceptibility to disturbances, and non-linear characteristics. In this regard, Machine Learning algorithms emerge as advanced techniques to increase the robustness to disturbances of model-based control systems. This work presents a model-free control framework based on a Deep Neural Network that autonomously determines load system parameters. A feedforward neural network has been conceptualized, designed, and constructed using a huge data set of a three-phase voltage source inverter driven by a Linear Algebra-based controller (LABC). Empirical findings reveal that Deep Adaptive Linear Based Algebra Control is able to reduce the Root Mean Square Error from ±1.68 A to ±0.29 A and achieve a notable enhancement in total harmonic distortion, with a reduction from 5.04% to 1.96% compared to LABC.