[Objective] This study addresses the limitations of conventional model predictive current control (MPCC) for doubly salient electromagnetic machine, including excessive current fluctuations and strong parameter dependency. An improved dual-vector model predictive current control (IDV-MPCC) strategy integrated with online parameter identification is proposed to enhance control robustness. [Methods] The proposed strategy achieved rapid sector localization of vectors, reduced the number of optimizations from 18 in traditional exhaustive search algorithms to just 4. This was accomplished by redefining the sectors and providing a new table for selecting vector combinations, which significantly reduced current prediction errors and ripples. To address parameter sensitivity issues, the model reference adaptive system (MRAS) was employed to perform online identification of the self-inductance and mutual inductance of the armature windings, and the identified results were fed back into the prediction model in real-time to correct it, thereby enhancing system robustness. [Results] The IDV-MPCC method significantly reduced current harmonics and torque ripples, while the MRAS-based identification effectively suppressed prediction errors caused by parameter mismatches. [Conclusion] The experimental results have verified that the proposed MRAS-IDV-MPCC method can effectively improve the control performance while reducing the computational burden of the system.