In this paper, a model predictive torque control method combining current reconstruction techniques for split-DC power converters is proposed for the torque ripple suppression of the switched reluctance machine (SRM). The proposed current reconstruction method does not require topology modification, and the phase currents are obtained through a combination of the collected neutral current and total currents according to the SRM operating region. This method does not require pulsation injection and therefore avoids the appearance of current distortion and loss of efficiency. Based on the reconstruction phase current, a new model predictive control (MPC) applied to split-DC converters is implemented to minimize the torque ripple of the SRM. Simulation results show that the proposed method can effectively reduce torque ripple with fewer power components and current sensors.