Abstract: Aiming at the characteristics of large inertia, long time delay, and large disturbance in the main steam temperature control system of power generation units, combined with the advantages of PI control and DMC, a PI-DMC based cascade feedforward predictive control strategy for the main steam temperature of power generation units was proposed. Using the control quantity as the fuel quantity and the controlled quantity as the main steam temperature, the forgetting factor recursive least squares method was used to identify the model parameters of the main steam temperature control system. By applying a unit step signal to the model, a predictive model was obtained, and the predictive value was corrected online through feedback correction. Real time rolling optimization of the objective function was performed to obtain the current control amount, and control weights were set to optimize the combination of PI and DMC. The simulation results show that compared with cascade PID control and DMC control, the proposed control strategy has a tuning time of about 20 s and an overshoot of only 6.47% when the model is adapted and interference is applied; When the model is mismatched and interference is applied, the adjustment time is about 30 s and the overshoot is 6.03%. The engineering application results show that compared to the original control strategy in the industrial field, the control accuracy of the proposed control strategy has increased by 66.91%, while the control accuracy of the drum liquid level has increased by 47.23%, which meets the on-site design requirements.