An adaptive residual-disturbance model predictive control strategy for dead-time compensation in PMSM drives

This paper proposes a Q-learning-based adaptive residual-disturbance model predictive control framework for permanent-magnet synchronous motor (PMSM) drives. The framework integrates an inverter dead-time voltage model, a dq -axis residual-disturbance extended-state observer, and model predictive control (MPC). The dead-time model provides feedforward compensation for inverter nonlinearities, whereas the observer estimates the remaining lumped disturbance. The disturbance estimate and its local variation trend are incorporated into the MPC prediction model. To improve adaptability under varying operating conditions, a Q-learning-based supervisory scheduler coordinates model-based compensation, disturbance estimation, and predictive optimization. It selects among bounded control modes according to zero-crossing proximity, current-tracking error, residual-disturbance intensity, and transient operating conditions. The reported simulation and experimental results show reduced zero-crossing current distortion and improved current and torque responses under the tested dead-time, load, and speed conditions.