Abstract: With the continuous tightening of air pollutant emission standards in China, traditional end-of-pipe control focusing on single pollutants can no longer meet the dual requirements of ultra-low emissions and cost control. Based on this, a multi-pollutant cooperative optimization system for typical large-scale industrial combustion sources was designed and implemented. The system integrated denitrification, dust removal and desulfurization units. A long short-term memory network was employed to develop multi-pollutant prediction models, and an operational cost model was established considering electricity and reagent consumption. Under emission and equipment constraints, particle swarm optimization and differential evolution algorithms were applied for global optimization. A case study on a power unit in Zhejiang province showed that the proposed method ensured stable compliance with SO2, NOx, and particulate matter standards, while reducing operational costs by 10%-20% on average across different load conditions. The results confirmed the engineering feasibility of the approach and highlighted its practical value for energy saving and emission reduction in large industrial combustion sources.

Key words: 　ultra-low emission, multi-pollutant cooperative optimization, LSTM prediction, evolutionary learning algorithm, cost minimization