Collaborative prediction and intelligent control of multiple pollutants emission from a large-scale waste incinerator

Abstract

Owing to the complexity of municipal solid waste (MSW), flue gas composition and operating conditions, it is challenging to predict pollutant emissions accurately and control them intelligently in the MSW incineration process. This study uses a 750 t/d large-scale grate-type MSW incinerator as the research object. Based on a long short-term memory (LSTM) model, collaborative prediction (co-prediction) of multiple pollutants (HCl, SO₂, NO_x, and PM) emissions from MSW incinerator flue gas was achieved. By coupling the prediction model with the particle swarm optimization (PSO) algorithm, an intelligent control program for pollutants developed with NO_x as an example can correlate NO_x emission with ammonia spray control. The results showed that, compared with conventional data input methods, time-series input resulted in better co-prediction performance. The mean absolute error (MAE) and mean squared error (MSE) results of the LSTM model on the testing set were reduced by 10.98% and 13.95%, respectively. The Change of MSE (COM) feature importance analysis method indicated that features such as the first flue temperature, the second flue temperature, and the primary air airflow had high importance in influencing the co-prediction of pollutants. The intelligent control program developed for NO_x emission was tested under continuous operation for 120 h, and compared with that achieved before optimization control, the amount of ammonia sprayed on the incinerator was reduced by 9.84% after optimization, reducing the environmental risk and offering significant economic benefits. This study provides scientific theoretical guidance for the efficient, economical and low-emission intelligent prediction and control of MSW incinerators.