Abstract: To improve the control performance of the coordinated control system in an ultra-supercritical thermal power unit, a control method was proposed based on iterative learning algorithm, which was used to improve the performance of the predictive controller. Based on the state space predictive model, iterative correction was made using historical errors. Meanwhile, the iterative sequence and time series quadratic terms were introduced into the performance indicators to calculate the optimal control law, thereby ensuring unbiased tracking of set values in case of model mismatch. For a 1 000 MW ultra-supercritical unit, an object model of superheated steam temperature was identified by particle swarm optimization algorithm based on field experimental data, while a simulation experiment was performed. Results show that the method proposed can effectively improve the control quality over superheated steam temperature, which has good resistance and robustness to I/O disturbance and model mismatch caused by wide load variations.