Optimized Model Predictive Control Using Machine Learning for HVDC Modular Multilevel Converters

The modular multilevel converter (MMC) is a highly promising converter for high-voltage direct current (HVDC) systems due to its modular design and ability to manage high power levels efficiently. However, its complex dynamics demand advanced control strategies to ensure optimal performance. Model Predictive Control (MPC) is particularly suitable for MMCs, as it utilizes mathematical models to predict future behaviors and optimize control actions. Despite its advantages, the computational burden of MPC can limit its application in real-time scenarios requiring rapid decision-making. To overcome this limitation, this study proposes an innovative framework that integrates neural network (NN)-based machine learning (ML) techniques into the MPC framework to enhance MMC control. Neural networks, including NN regression, NN Pattern Recognition (NNPR) for regression, and NN Function Fitting (NNFF), are trained using comprehensive simulation datasets to predict optimal control actions under various conditions precisely. Simulation results show a substantial reduction in Total Harmonic Distortion (THD), illustrating the machine learning-enhanced MPC's ability to decrease computational burden while ensuring reliable MMC operation. Comparative analysis shows that NNPR for regression is the most efficient.