A Novel Capacitance Estimation Method of Modular Multilevel Converters for Motor Drives Using Recurrent Neural Networks with Long Short-Term Memory

Abstract

Accurate estimation of submodule capacitance in modular multilevel converters (MMCs) is essential for optimal performance and reliability, particularly in motor drive applications such as permanent magnet synchronous motor (PMSM) drives. This paper presents a novel approach utilizing recurrent neural networks with long short-term memory (RNN–LSTM) to precisely estimate capacitance in MMC-based PMSM drives. By leveraging simulation data from MATLAB, the LSTM neural network is trained to predict capacitance based on voltage, current, and their temporal variations. The proposed LSTM architecture effectively captures the dynamic behavior of MMCs in PMSM drives, providing high-precision capacitance estimates. The results demonstrate significant improvements in estimation accuracy, validated through mean squared error (MSE) metrics and comparative analysis of actual versus estimated capacitance. The method’s robustness is further confirmed under varying operating conditions, highlighting its practical utility for real-time applications in power electronic systems.