Time-delay is widely found in industrial process systems, and traditional controllers are not competent to deal with time-delay systems as the difficulty of precise modeling. For this type of system, combined with tracking-differentiator for obtaining differential information, an improved model-free adaptive predictive control (IMFAPC) method is proposed by using Smith predictor to predict future output. The convergence and stability of the method are analyzed. This method is a data-driven control method which does not require the information of the controlled system, only requires real-time I/O data. Numerical simulation and simulation of the outlet temperature control of cement calciner show the excellent performance and faster response of proposed method, and its effectiveness is verified.