During the operation of thermal power units, a large amount of NOx pollutants is generated due to combustion. In order to reduce NOx emissions, the implementation and optimization of denitrification technology has become a key issue in the current environmental protection. In the selective non-catalytic reduction (SNCR) denitration control system, due to the delay in the chemical reaction time of sampling, it is impossible to accurately predict the NOx concentration and to quickly adjust the amount of ammonia injection. In this work builds a SNCR denitrification model based on Bi-LSTM to predict the main steam flow rate and the flow rate of urea solution in the furnace area at the current moment using upstream boiler operation parameters. Comparing with SVR, DNN, CNN, LSTM network, the proposed method performs the best in simulation. The MSEs of main steam flow and furnace zone urea solution urea flow are as low as 0.000151 and 0.001232, respectively, and the R2 s are as high as 0.998905 and 0.998905, respectively.