Mathematical and Artificial Neural Network Modeling and Simulation of NOx Selective Catalytic Reduction in Catalytic Bed

Worldwide development of industries and increasing of energy consumption, have resulted increasing of the emission of nitrogen oxide, NOx, pollutants produced by fossil fuels in industries and internal combustion engines. Therefore, NOx emission control and its removal is very important. In this study, modeling and simulation of selective catalytic reduction (SCR) of NOx in a catalytic bed at both steady-state and dynamic conditions was performed. Results of steady-state simulation showed that because of the intense effect of temperature on NOx conversion and competition of the main reaction with the oxidation of NH3 by O2, conversion of NOx requires a catalytic filter in the range of 300-350°C . The results showed that NO conversion increases with decreasing gas hourly space velocity (GHSV) and increasing inlet NO concentration. At dynamic condition, steady-state results were used as initial conditions for dynamic simulation and the effect of changes of the effective parameters including GHSV, NO concentration, and NH3/NO ratio were investigated. Also, steady-state simulation of the process were performed using a feed-forward artificial neural network and conversion values of NOx and NH3 were estimated as a function of GHSV, reactor temperature, and NO concentration at fixed NH3/No ratio. 96 networks with different neurons and two different activation function in hidden layer were trained three times with different initial weights. The resulted optimum network showed maximum mean square of errors about 0.01 with respect to mathematical modeling results indicating high performance of neural network for predication of process performance.