An NARX-neural-network with external input was proposed based on mutual information(MI) variable selection. A dynamic model for selective catalytic reduction(SCR) of flue gas denitrification system was established. The factors influencing the SCR flue gas denitration system were analyzed by the mutual information. The particle swarm optimization(PSO) algorithm was applied to select the system input variables and ascertain the time delay of the input parameters. On this basis

the selected input variables were applied to the NARX neural network

and the dynamic model of the SCR system was established. The operation data of the SCR denitration system of a coal-fired unit verified that the model has good prediction accuracy and generalization ability.