It is difficult to accurately measure the NO_x mass concentration at inlet of the SCR denitrification system of coal-fired units. To solve this problem

a prediction model based on feature optimization and improved long short-term memory(LSTM) network is proposed. Firstly

auxiliary variables related to NO_x emissions are determined by mechanism analysis

and the delay time between auxiliary variables and output variables is calculated by mutual information. Then

the maximum correlation and minimum redundancy(mRMR) algorithm is used to determine the optimal feature subset of the auxiliary variables. The RBF network is used to compare the prediction errors of each feature subset to determine the model input variables. Finally

through grid search and modified particle swarm optimization(MPSO) algorithm

the super parameters of the LSTM network are determined

and the prediction model for NO_x mass concentration is established. The simulation results show that

the auxiliary variables after feature optimization improve the generalization ability of the model

and the model's prediction accuracy is improved by using the LSTM network with memory function.