Boiler NOx emission prediction based on ensemble learning and extreme learning machine optimization

The nitrogen oxides (NOx) emission measurement of selective catalytic reduction (SCR) denitrification system has issues that insufficient live processing and irregular purge readings. Therefore, establishing an accurate NOx concentration prediction model can significantly advance the timeliness and precision of NOx measurement. The study proposes a prediction method based on ensemble learning and extreme learning machine (ELM) optimization to build a NOx concentration prediction model for SCR denitrification system outlet. Firstly, to enhance the modeling precision of ELM for complex feature objects under all working conditions, the ensemble learning framework was introduced and an ensemble learning model based on ELM was designed. Secondly, to alleviate the impact of random initialization of ELM network learning parameters on the stability of modeling performance, the multi strategy improved dingo optimization algorithm (MS-DOA) is given by introducing Tent chaotic mapping, Lévy flight and adaptive t-distribution strategy to ameliorate the initial solution and position update process of population. Finally, the SCR denitrification operating data from 660 MW coal-fired power plant was opted for experimental validation. The findings demonstrate that the established SCR denitrification system outlet NOx concentration prediction model has high modeling accuracy and prediction accuracy, and provides a reliable approach for achieving accurate prediction of boiler NOx emissions.