Reinforcement learning-driven model predictive control for optimizing counter-rotating permanent magnet synchronous motor in submarine propulsion system

Counter Rotating Permanent Magnet Synchronous Motors (CRPMSM) are increasingly favored in underwater application due to their high torque density, efficiency and ability to cancel out yaw inducing moments through the use of dual rotors spinning in opposite directions. However, ensuring synchronization between the rotors under varying load dynamic underwater conditions poses significant control challenges. To address these limitation this research proposed a Reinforcement Learning-Driven Model Predictive Control (RL-MPC) for optimizing the performance of CRPMSM in submarine propulsion systems. RL-MPC control architecture used a Twin Delayed Deep Deterministic Policy Gradient (TD3) reinforcement learning. The system is modeled in MATLAB/simulink with CRPMSM represented in d-q reference frame and driven by voltage source inverter (VSI) using Space Vector Pulse Width Modulation (SVPWM). The RL-MPC controller performance evaluated under three condition: constant speed with variable balanced load, variable speed with constant load and constant speed with unbalanced load variation. Simulation result confirm that the RL-MPC improves motor performance by enhancing speed tracking, reducing torque ripple, maintaining rotor synchronization improving transient response compared to standalone MPC. Quantitative comparison shows RL-MPC enhances dynamic performance comparatively over single MPC. The total harmonic distortion (THD) of stator current during unbalanced load resynchronization was enhanced from 9.3% (MPC) to 3.4% (RL-MPC), overshoot decreased from 30% to 16.6%, and settling time was enhanced from 1.4 s to 0.7 s. These enhancements validate RL-MPC achieves a 63.4% reduction in THD, 45% reduction in overshoot, and 50% enhancement in settling time under unbalanced load conditions. Finally the Lyapunov-based stability analysis confirms the closed-loop stability of the system.