火电机组参与灵活调峰运行时，负荷的大范围变动会引起煤量、风量等参数频繁变化，进而导致主汽温剧烈波动，当前汽温控制系统难以适应灵活调峰工况下的需求，发展针对性的优化控制策略，对提高主汽温的控制品质、维持机组安全稳定运行至关重要。以某600MW汽包锅炉为研究对象，提出一种基于神经网络预测的主汽温控制优化方法，其基于随机森林算法挖掘出对主汽温起主要影响的因子；针对各影响因子存在的迟延问题，提出基于互信息法的迟延估计方法，实现神经网络输入参数的时延对齐优化；采用CNN-LSTM-Attention神经网络构建主汽温预测模型，实现对未来时刻主汽温的变化进行预测，并构建主汽温的预估补偿系统。仿真结果表明，此方法可以有效提升主汽温的预测精度，降低负荷频繁扰动下主汽温的波动幅度，提前抑制各扰动量对蒸汽温度造成的影响。

When thermal power units participate in flexible peak shaving operation, large-scale changes in load can cause frequent changes in parameters such as coal quantity and air volume, leading to severe fluctuations in main steam temperature. The current steam temperature control system is difficult to adapt to the needs of flexible peak shaving conditions. Developing targeted optimization control strategies is crucial for improving the control quality of main steam temperature and maintaining safe and stable operation of the unit. A main steam temperature control optimization method based on neural network prediction is proposed, taking a 600MW drum boiler as the research object. The method uses a random forest algorithm to mine the factors that have a major impact on the main steam temperature; Aiming at the delay problem of various influencing factors, a delay estimation method based on mutual information method is proposed to achieve delay alignment optimization of neural network input parameters; Using CNN-LSTM Attention neural network to construct a main steam temperature prediction model, to predict the changes in the main steam temperature at future times, and to construct an estimated compensation system for the main steam temperature. The simulation results show that this method can effectively improve the prediction accuracy of main steam temperature, Reduce the fluctuation amplitude of main steam temperature under frequent load disturbances, and suppress the impact of various disturbances on steam temperature in advance.

国家自然科学基金重点项目（61833011）