Mechanism-enhanced data-driven method for the joint optimization of boiler combustion and selective catalytic reduction systems considering gas temperature deviations

Abstract

A novel optimization framework for boiler combustion and selective catalytic reduction systems (BCSCRSs) that accounts for flue gas temperature deviation and integrates prior knowledge is proposed. Initially, single-factor hot-state experiments on the BCSCRSs are conducted to yield samples for model training. By integrating monotonic relationships from the experimental data into the fusion monotony–support vector regression model, more reasonable estimates and trends of the prediction targets (e.g., CO and NOx concentrations, and unburned combustibles) are obtained compared to those of the model without prior knowledge. Based on this model, the inequality-constrained reference vector-guided evolutionary algorithm is employed for multiobjective combustion optimization considering the constraints of gas temperature deviation in the horizontal pass. The corresponding Pareto-frontier solutions that serve as boundary conditions for the second-stage optimization are obtained. The second-stage optimization comprehensively considers the total operating cost of the BCSCRSs, the denitrification fraction in the NOx formation and removal processes, and safety concerns related to steam tube overheating and ammonia slip. Several noteworthy results are derived: the adjustment of the burner tilt angle presents a trade-off between eliminating gas temperature deviation and reducing NOx formation. An optimal point can be identified by comparing the magnitudes and variations of different BCSCRSs objectives.