考虑超临界锅炉过热汽温系统的非线性、大惯性、大时延等特性，建立了过热汽温喷水减温系统的非线性动态神经网络逆模型，运用机组的历史运行数据对模型进行训练与校验。以训练好的模型为基础，构建了具有PID补偿环节的神经网络逆控制器，在MATLAB平台编制了实时控制程序。借助600MW超临界机组全仿真系统，对过热汽温进行设定值扰动、大范围变工况扰动等仿真试验。结果表明：具有PID补偿的神经网络逆控制方案可有效降低动态变负荷过程中过热汽温的控制偏差，缩短汽温控制的稳定时间，与机组原控制方案相比具有更好的控制效果。

Considering the characteristics of superheated steam temperature (SST) system of supercritical boiler such as nonlinearity, large inertia and long time delay, the nonlinear dynamic neural network inverse model of superheated steam temperature spray desuperheating system was established. It was trained and verified by the historical operation data of the unit. Based on the trained model, the neural network inverse controller with PID compensation link was constructed, and the real-time control program was programmed on the MATLAB. The whole simulation system of 600MW supercritical unit was used to simulate the temperature of superheated steam, such as the disturbance of set point and the disturbance of large scale off-design condition. The result shows that the neural network inverse control scheme with PID compensation can effectively reduce the control deviation of the superheated steam temperature and shorten the stable time of the steam temperature control during the dynamic load changing process, and has better control effect than the original control scheme of the unit.

国家自然科学基金项目（面上项目，重点项目，重大项目）