Intelligent air flow control in municipal solid waste incineration power plants: a novel approach integrating machine learning and physical-chemical models

Accurate airflow control in Municipal Solid Waste Incineration (MSWI) is essential for stable combustion, low emissions, and high operational efficiency. Despite advancements in control strategies, existing methods often fail to adapt to dynamic operational conditions and environmental fluctuations, leading to suboptimal performance. In this paper, an integrated control approach, combining physical-chemical modeling with machine learning for real-time prediction and regulation of primary and secondary air flows, was proposed. The integrated model leverages historical data and dynamically responds to operational variables, enhancing adaptive air flow regulation. Firstly, a dynamic physical-chemical model was developed to simulate mass transfer and combustion reactions. Subsequently, a Convolutional Neural Network - Long Short-Term Memory model (CNN-LSTM) optimized by the Gray Wolf Optimizer (CNN-LSTM-GWO) significantly improved airflow prediction accuracy (R2 = 0.8724, RMSE = 3.79 %). Specifically, the RMSE for primary airflow prediction was reduced by 2.21 % compared to GA and 2.16 % compared to PSO, while for secondary airflow, the reductions were 2.04 % and 1.86 %, respectively. The integrated control model adjusts grate speed, air ratio, and excess air coefficient to achieve precise airflow distribution. Results demonstrated that the proposed method accurately predicted key MSWI outputs (furnace temperature, steam flow, oxygen concentration) with less than 5 % error and effectively reduced NOx and CO emissions, highlighting its potential for widespread application in industrial settings.