Abstract: Affected by the sampling system and measuring instruments, it is difficult to ensure the automatic input effect of the traditional ammonia injection control. After the automatic input, the NOx at the denitration outlet fluctuates greatly. The reasons are: the controlled object (denitration device) has the characteristics of large lag, nonlinear and timely variability, and the flue gas sampling analysis is not representative. In order to solve the above problems, an advanced SCR intelligent ammonia injection control strategy based on closed-loop decoupling is proposed in a plant reconstruction project to replace the traditional PID control. It uses neural network technology to deal with the nonlinear and timely variability of the controlled process, and through neural network soft measurement technology, online evaluation of measurement parameters, and variable structure control technology to ensure that the control system can be automatically put into operation even if some parameters are distorted or in maintenance. This control strategy has positive reference significance for ammonia injection optimization of similar units.