During the switching process of permanent magnet motors， the current fluctuates intensely and is difficult to predict due to the coupling effect of multiple factors such as internal electromagnetic fields， nonlinear dynamic characteristics， and external loads. To enhance the operational stability of the motor， this paper proposes a switching current prediction method based on fractional-order neural networks. The research first constructs a set of equations covering voltage， flux linkage， and torque to accurately describe the dynamic characteristics of the motor. Then， the least squares method is used for iterative optimization to achieve high-precision identification of model parameters. The deviation of the inductance value obtained through identification from the real-time measured value is obtained and included in the input of the fractional-order neural network. This network uses the theory of fractional calculus to describe the dynamics of neurons， accurately captures the current change patterns， and achieves reliable prediction. Experiments were conducted under the condition that the load increased from 25% to 175%. The results show that the maximum oscillation amplitude during prediction is only 0.5%， demonstrating excellent prediction accuracy and stability， effectively reducing current fluctuations， and providing an important technical guarantee for the stable operation of the motor control system.