Abstract

The Modular Multilevel Converter (MMC) has emerged as the preferred architecture for integrating renewable energy power plants into the grid via undersea HVDC cables and HVDC overhead lines. However, to mitigate the significant power pulsations introduced by the single-phase AC–DC conversion in its arms, the MMC necessitates the use of large DC-link capacitors. To address these limitations, the Alternate Arm Converter (AAC) topology integrates MMC arms with Director Switches (DS), thereby reducing the required number of sub-modules, potentially by half, when compared to conventional MMC configurations. Furthermore, the AAC provides inherent DC fault-blocking capability, which is a critical feature for future DC grid applications. Model Predictive Control (MPC) is widely employed due to its flexibility in incorporating multiple control objectives within a unified cost function. Nonetheless, the extensive number of possible switching states in AAC results in substantial computational complexity, posing challenges for real-time control implementation. To alleviate this computational burden, this work proposes the development of machine learning (ML)-based controllers for AAC, leveraging training data generated through model predictive control (MPC) operation. The system architecture utilizes a solar photovoltaic (PV) array as the DC energy source, interfaced with the AC grid via the proposed AAC-based high-voltage direct current (HVDC) transmission link. In this study, an artificial neural network (ANN) is employed as the machine learning (ML) framework. The wellness of the proposed scheme is confirmed over a comprehensive simulation analysis that employs a detailed switching converter model.

References

Citation

Copy

Copyright

Article metrics