A Deep LSTM-Based RSF Capacitor Voltage Balancing Method for MMC

The article suggests a novel approach for balancing the capacitor voltage in modular multilevel converters (MMCs) using a deep Long Short-Term Memory (D-LSTM) based technique. The D-LSTM network model is trained and tested offline using data collected from the Reduce Switching Frequency (RSF) capacitor voltage balancing method. This trained model is then used to control the MMCs accurately. By shifting most of the computational burden from online to offline, the proposed D-LSTM model can replace the RSF capacitor balancing method with a significantly more computationally efficient structure, which can handle large-scale datasets without compromising accuracy. Simulation results demonstrate that the proposed D-LSTM model performs just as well as the capacitor balancing method, with a test set accuracy of 99.4 %. This finding has broad implications for using the D-LSTM method for complex controllers in power electronic systems.