Sensorless Control of Permanent Magnet Synchronous Motors Using LSTM-Enhanced Extended Kalman Filtering

Sensorless control of surface-mounted permanent magnet synchronous motors (SPMSMs) under low-speed and heavy-load conditions represents a pivotal research topic in motor control. The conventional high-frequency injection method estimates rotor position by analyzing high-frequency current responses in the d-q axes. However, under heavy-load conditions, magnetic saturation in the iron core causes a reduction in d-q axis inductances and a decreased inductance differential, significantly impairing the observability of the electrical model. Although mechanical models can provide position and speed information, their performance is constrained by nonlinear mechanical parameters and uncertainties. To address the challenges of magnetic saturation and mechanical parameter uncertainties, this paper proposes a sensorless control strategy for electromechanical models that integrates a long short-term memory (LSTM) network with an extended Kalman filter (EKF). This strategy leverages the LSTM to capture the dynamic characteristics of inductance as well as mechanical parameter uncertainties, enabling real-time optimization of the EKF’s noise covariance matrix. By establishing a closed-loop compensation mechanism that correlates "inductance variations and parameter uncertainties → statistical properties of observation noise → Kalman gain, " the proposed approach substantially enhances the accuracy of rotor position estimation. Simulation results demonstrate that this scheme significantly improves control robustness and estimation precision under low-speed and heavy-load conditions.