With the increase in the proportion of new energy power generation in grids, there is an urgent need for the flexible operation of coal‐fired power plants. However, as the most efficient commercial coal‐fired power generation technology, the flexibility of double‐reheat ultra‐supercritical units is limited due to the steam temperature fluctuation in transient processes. To inhibit the steam temperature fluctuations, a data‐driven method, called long short‐term memory (LSTM) was introduced to model the dynamics of the steam temperatures of a 1000‐MW double‐reheat coal‐fired boiler. Moreover, an advanced controller was proposed by combining the LSTM‐based dynamic model and the model predictive control architecture. The results demonstrate that the LSTM model can be used to capture the time delay, multivariable coupling and non‐linear dynamics of the steam temperatures and the proposed controller exhibits excellent performance in steam temperature regulation. The superheated steam temperature fluctuation decreases from 15.4 to 9.8°C, the primary reheat temperature fluctuation decreases from 34.6 to 8.3°C, and the secondary reheat temperature fluctuation decreases from 41.4 to 8.8°C. The proposed method has good potential to stabilize steam temperatures during transient processes of wide‐range load cycling, and finally improves the flexibility of the double‐reheat ultra‐supercritical unit.