Aiming at the problems of unrecovered energy during the braking process of electric-wheel mining trucks and insufficient robustness of motor dynamic control, this study designs an optimized motor control method based on model predictive torque control. By combining a dynamic state-of-charge management model for supercapacitors with the discretized state-space model of a three-phase asynchronous motor in the α-β coordinate system, a multi-objective cost function is constructed that includes torque tracking error, flux linkage amplitude constraints, and inverter switching frequency. Additionally, a one-step delay compensation mechanism is introduced to predict system states in advance and adjust control inputs, effectively suppressing torque ripple in traditional finite control set model predictive control. Using MATLAB/Simulink, a braking energy feedback control model for electric-wheel mining trucks is established. The experimental results show that under load mutation conditions, the torque ripple is reduced from 30 in PI control to 20; under speed mutation conditions, the system response time is less than 0.1 s, and the total harmonic distortion of the current decreases from 5.2% to 2.1%.