Abstract: When parameter mismatches, the explicit model predictive (EMP) direct speed control of permanent magnet synchronous motor will have obvious steady-state deviations. To this end, the existing method configures an extended state observer (ESO) to observe and feed-forward compensate for model deviations in real-time, thereby achieving offset-free and high-precision speed tracking control. However, experimental and theoretical analyses have found that the system's dynamic response deteriorates when parameter mismatches due to the ESO's limited bandwidth, which leads to a weak ability to compensate for changing disturbances. To simultaneously improve the system's steady-state control accuracy and dynamic performance with parameter mismatches, as well as enhance robustness, this paper integrates model-free control with EMP control and proposes a new EMP direct speed control strategy by constructing an ultra-local predictive model and a data-driven observer. Experimental results show that the proposed method can achieve excellent compensation for parameter mismatches in both dynamic and steady-state stages by virtue of the data-driven observer's high observation bandwidth, thereby ensuring both dynamic and steady-state performance.