Investigation on the dual-stage SCR system control strategy based on NOx and exhaust temperature prediction model

Abstract

With the advancements in deep learning and neural network technologies within the engine technology domain, intelligent control of exhaust after-treatment systems has become viable. The exhaust temperature significantly affects the NOx conversion efficiency of the SCR system. In this study, a time-series prediction model for exhaust temperature and NOx emissions was first developed using the LSTM neural network based on experimental data, and a Multi-Head Attention Mechanism was introduced to enhance the model’s predictive performance (LSTM-MA model). Subsequently, a dual-stage SCR system’s ammonia-to-NOx ratio control method (LSTM-ANR) was developed based on LSTM-MA model. The potential of the dual-stage SCR system using the LSTM-ANR control method to reduce NOx emissions during the WHTC cycle was analyzed using GT-Power and Simulink software. Finally, the experimental results show that the cold-state and hot-state WHTC cycle’ weighted NOx emission and conversion efficiency are 0.165 g/kW·h and 98.6%, respectively. Compared to the original data from the engine, NOx emission was reduced by 64.13%. This proves that the control method can effectively lower NOx emission.