Direct predictive speed control (DPSC) is an appropriate solution to achieve high-performance cascade-free control structure of permanent magnet motors, which has the technical limitations of large amount of online calculation and limited control degree of freedom. Based on the two-level inverter-fed permanent magnet synchronous motors (PMSMs) drive system, this paper proposes an improved DPSC strategy with incremental predictive model for improving the speed dynamic control performance of the system. First, in order to reduce the amount of online calculation of the algorithm, a low-complexity incremental speed prediction model was constructed, which can realize long-horizon multi-step speed prediction and ensure the stability of the algorithm. Then, the control-set was expanded with virtual vectors to increase the control degree of freedom, and its corresponding vector synthesis method and two-stage exhaustive optimization were designed to screen the optimal output vector efficiently. Combined with the established speed prediction model and control-set, the stable speed control performance could be obtained with relatively low computational burden. Experimental results show that compared with classical speed control strategies, the proposed strategy had better speed dynamic regulation ability, and the speed reached its reference value rapidly and directly without shooting. Besides, the proposed strategy could operate normally under the condition of parameter mismatch.