Data-driven evaluation and prediction of the gas-fired boiler performance under the impact of fine desulphurization gas

Abstract

Fine desulfurization gas (FDG), as the waste gas produced during the regeneration of desulfurization agents in coke oven gas desulfurization systems, can be blended with blast furnace gas and fed into gas-fired boilers for consumption. However, the impact of FDG incorporation on the boiler performance is still not clear to date. In this work, by examining the fuel and gas samples after FDG incorporation, the concentration of containments such as H₂S and CS₂ can be observed significantly to increase during the hot blow stage. Using the mutual information method, it can be found that the boiler efficiency and the SO₂ emission are closely related to FDG. With increasing FDG temperature and flow rate, there is a certain linear improvement in the boiler efficiency, but accompanied by a significant increase in the SO₂ concentration. To overcome this issue, prediction models for SO₂ concentration were comparatively developed based on the light gradient boosting machine (LightGBM) algorithm and the long-short term memory network, respectively. The results show that the LightGBM model combined with the auto-regression features can achieve an accurate prediction of the SO₂ concentration after FDG incorporation, with a root-mean-squared error of only 0.09 mg/m³. The current work not only provides reliable evidence for the changes in gas-fired boiler performance after the incorporation of gases containing multiple containments, but also offers a new orientation for the efficient utilization of waste gases in the relevant industries.