Q-Learning-Based Smart Selective Harmonic Current Mitigation-PWM (S2HCM-PWM) for Grid-Connected Converters

Abstract

Multilevel converters become more and more interesting for renewable energies and energy storage systems. Various modulation techniques such as high-frequency modulation approaches (e.g., space vector modulation and phase shift-PWM) and low-frequency modulation approaches (e.g. selective harmonic current mitigation-PWM (SHCM-PWM), selective harmonic mitigation-PWM (SHM-PWM), and selective harmonic elimination-PWM (SHE-PWM)) are employed for multilevel grid connected converters in the literature. High efficiency (low switching losses) can be achieved by using the low-frequency modulation approaches. However, low-frequency modulation techniques significantly increase the coupling inductance (passive filter). High-switching frequency modulation techniques have a better dynamic response and use a smaller passive filter. In this paper, a machine learning technique (Q-learning) is used to have advantages of both high- and low-frequency modulation approaches. The proposed smart modulation technique meets all current harmonic requirements, while the switching frequency of the converter is not significantly increased. To evaluate the effectiveness of the proposed technique, simulations are conducted on a 7-level (3-cell) single-phase cascaded H-bridge converter.