Deep Learning-based Control Perspective for Single-Phase Grid-connected Inverter Using Gated Recurrent Units

Abstract

This paper introduces a novel approach to control single-phase grid-connected inverters (GCIs) using artificial intelligence (AI), specifically employing a deep learning- based method with Gated Recurrent Unit (GRU) networks. The proposed GRU-based controller is trained offline using TensorFlow and Keras libraries in Python, and is subsequently implemented for real-time applications. Comparative analysis between the GRU-based controller and the conventional PI controller reveals distinct advantages of the former, including improved transient response and reduced oscillations. Furthermore, the GRU-based controller demonstrates superior performance, reducing the total harmonic distortion (THD) and efficiently regulating current in the presence of varying grid conditions.