Abstract: NO_x mass concentration prediction and ammonia injection optimization are of great significance for SCR system and unit operation. In order to overcome the problem of different time delays and mutual coupling among various operating variables of the SCR system, a model for the prediction of NO_x mass concentration and optimization of ammonia injection volume considering time delays is proposed. Firstly, the delay time between variables is determined according to the maximum mutual information coefficient between variables, and the data set is reconstructed. The optimal input variable set is determined based on the mean influence value method. And then the neural network prediction model of NO_x mass concentration for SCR system outlet is established. Finally, the particle swarm optimization algorithm is used to optimize the ammonia injection amount of the SCR system at the outlet, so as to reduce the ammonia escape amount at the outlet as much as possible under the premise of avoiding NO_x exceeding the standard. The test results based on the actual operation data of the power plant show that the proposed model can better predict the outlet NO_x mass concentration. Considering the time delay of each variable can improve the prediction accuracy of SCR system. The mean influence value method for feature processing can greatly reduce the training time, and has little impact on the accuracy of the model; Particle swarm optimization can be used to guide the adjustment of ammonia injection amount and avoid the outlet NO_x mass concentration exceeding the standard.