Recurrent Neural Network-Based Fast Adaptive Control for Smooth Speed Regulation of PMSMs

High-precision smooth speed regulation in permanent magnet synchronous motors (PMSMs) is challenged by torque ripple, parameter mismatches, and load fluctuations. Considering the uncertain nature of these disturbances, neural network (NN)-based methods provide superior applicability compared to internal model approaches due to their frequency-independent characteristics. However, most of these methods utilize feedforward NNs designed via asymptotic stability theory, which exhibit severe output distortions stemming from abrupt error surges and prolonged convergence times under varying operating conditions. To address these issues, this article proposes a robust speed control framework that integrates a recurrent NN (RNN) with fixed-time sliding-mode control (FTSMC), theoretically guaranteeing closed-loop fixed-time stability even under input saturation. Specifically, the RNN employs an internal recurrent loop to leverage historical context, thereby reducing the network output's susceptibility to instantaneous error spikes and alleviating transient distortions. Furthermore, incorporating FTSMC principles into the weight update law design accelerates the network's learning process, ensuring accurate approximation of periodic disturbances while shortening the transient period. Extensive experimental results validate the effectiveness of the proposed scheme across diverse operating scenarios.