Academic Journal of Materials & Chemistry, 2025, 6(3); doi: 10.25236/AJMC.2025.060315.
Bingyi Guo1, Wenjie Zhang1, Yunjian Duan1, Jizhou Ding1, Linjie Li1, Ayizeba Maimaituxun2
1School of Applied Chemical Engineering, Lanzhou Petrochemical University of Vocational Technology, Lanzhou, Gansu, China
2School of Mathematics and Statistics, Northwest Normal University, Lanzhou, Gansu, China
Flue gas desulfurization (FGD) is a key unit operation in waste-to-energy plants, directly affecting air-pollutant emissions and process safety. However, complex reaction–absorption mechanisms, fluctuating waste composition and coupled operating variables make it difficult to maintain outlet SO₂ and H₂S concentrations within specification using conventional empirical correlations. This paper proposes a data-driven framework that integrates static process modeling, global sensitivity analysis and dynamic early-warning prediction for an industrial FGD system. First, multiple regression, ensemble learning and nonlinear machine-learning models are trained on historical operation data to map key process variables (e.g., slurry pH, liquid-to-gas ratio, oxidation–reduction potential, temperature and flow rates) to outlet SO₂ and H₂S concentrations. Model comparison shows that tree-based ensemble models achieve high accuracy (R² > 0.97) and are selected as surrogate process models. Second, Sobol global sensitivity analysis is applied to quantify the contribution of individual variables and their interactions, revealing that slurry pH, oxidation–reduction potential and absorber temperature are the dominant factors governing desulfurization performance. Third, a hybrid LSTM–ARIMA residual model is developed to predict future outlet concentrations and derive a binary early-warning signal for potential non-compliance. A multi-layer adaptive threshold optimization framework is introduced to tune the decision threshold by jointly considering prediction uncertainty and the trade-off between missed alarms and false positives. Finally, a multi-scale time-localization strategy refines the predicted onset time of non-compliant events within a moving risk window. Case-study results on plant-scale data demonstrate that the proposed approach can achieve over 90% classification accuracy for prediction horizons up to 40 time steps, with an average timing error of fewer than two sampling intervals. The framework provides a practical tool to support proactive operation, tighten emission control and enhance process safety in industrial FGD units.
Flue Gas Desulfurization; Waste-to-Energy Plant; Machine Learning; LSTM–ARIMA Hybrid Model; Sobol Sensitivity Analysis; Early Warning; Process Safety
Bingyi Guo, Wenjie Zhang, Yunjian Duan, Jizhou Ding, Linjie Li, Ayizeba Maimaituxun. Data-Driven Modeling, Sensitivity Analysis and Early Warning of Flue Gas Desulfurization in Waste-to-Energy Plants. Academic Journal of Materials & Chemistry (2025), Vol. 6, Issue 3: 108-119. https://doi.org/10.25236/AJMC.2025.060315.
[1] Naderi, K., Kalami Yazdi, M.S., Jafarabadi, H., Bahmanzadegan, F., Ghaemi, A. and Mosavi, M.R. Modeling based on machine learning to investigate flue gas desulfurization performance by calcium silicate absorbent in a sand bed reactor. Scientific Reports, 2024(14): 954. DOI: 10.1038/s41598-024-51586-7.
[2] Huang, J., Zeng, Z., Hong, F., Yang, Q., Wu, F. and Peng, S. Sustainable operation strategy for wet flue gas desulfurization at a coal-fired power plant via an improved many-objective optimization. Sustainability, 2024,16(19): 8521. DOI: 10.3390/su16198521.
[3] Liu, Q., Li, X. and Wang, K. Dynamic modeling of flue gas desulfurization process via bivariate EMD-based temporal convolutional network. Applied Sciences, 2023, 13(13):7370. DOI: 10. 3390/app13137370.
[4] Li, R., Zeng, D., Li, T., Ti, B. and Hu, Y. Real-time prediction of SO₂ emission concentration under wide range of variable loads by convolution–LSTM VE-transformer. Energy, 2023(191):126781. DOI: 10.1016/j.energy.2023.126781.
[5] van den Boorn, B.F.H., van Berkel, M. and Bieberle-Hütter, A. Variance-based global sensitivity analysis: A methodological framework and case study for microkinetic modeling. Advanced Theory and Simulations, 2023,6(10):2200615. DOI: 10.1002/adts.202200615.
[6] Ju, J., Liu, K. and Liu, F. Prediction of SO₂ concentration based on AR-LSTM neural network. Neural Processing Letters, 2023(55):5923–5941. DOI: 10.1007/s11063-022-11119-7.
[7] Qiao, Z., Wang, X., Gu, H., Tang, Y., Si, F., Romero, C.E. and Yao, X. An investigation on data mining and operating optimization for wet flue gas desulfurization systems. Fuel, 2019(258):116178. DOI: 10.1016/j.fuel.2019.116178.
[8] Guo, Y., Xu, Z., Zheng, C., Shu, J. and Dong, H. Modeling and optimization of wet flue gas desulfurization system based on a hybrid modeling method. Journal of the Air & Waste Management Association, 2019,69(5):565–575. DOI: 10.1080/10962247.2018.1551252.
[9] Li, C., Zhang, B., Li, J. and Hu, Y. Process simulation of wet flue gas desulfurization. Journal of Chemical Engineering Research Updates, 2024,6(1), 1–10. DOI: 10.15377/2409-983X.2024.11.5.
[10] Dong, Q., Wang, C., Peng, S., Wang, Z. and Liu, C. A many-objective optimization for an eco-efficient flue gas desulfurization process using a surrogate-assisted evolutionary algorithm. Sustainability, 2021, 13(16): 9015. DOI: 10.3390/su13169015.
[11] Riaz, F., Awan, M.R., Nabi, H.Z., et al. A machine learning based multi-objective optimization for flue gas desulfurization enhancement in coal power plants. Energy Nexus, 2025(20): 100534. DOI: 10.1016/j.nexus.2025.100534.
[12] Makomere, R., Rutto, H. and Koech, L. The assessment of response surface methodology (RSM) and artificial neural network (ANN) modeling in dry flue gas desulfurization at low temperatures. Journal of Environmental Science and Health, Part A, 2023,58(3):191–203. DOI: 10.1080/10934529. 2023. 2174334.
[13] Makomere, R.S, Chacha, J.S., Mkilima, T. and Onyango, M.S. Precision forecasting of spray-dry desulfurization using artificial neural networks. Journal of Environmental Science and Health, Part A, 2024,59(4):321–333. DOI: 10.1080/10934529.2024.2317670.
[14] Shayanmehr, M., Keshavarz Moraveji, M. and Behbahani, R. A data driven machine learning approach for predicting and optimizing sulfur compound adsorption on metal–organic frameworks. Scientific Reports, 2025(15):3138. DOI: 10.1038/s41598-025-86689-2.
[15] Yin, X., Li, J., Wu, Y., et al. Enhancing deep learning for the comprehensive forecast of SO₂ emission in a wet flue gas desulfurization system. Control Engineering Practice, 2023(139): 105587. DOI: 10.1016/j.conengprac.2023.105587.
[16] Liu, Q., Li, X. and Wang, K. Real-time dynamic modelling of industrial WFGD process via multi-model and neural network approaches. Measurement, 2025(249): 117051. DOI: 10.1016/j. measurement. 2025.117051.
[17] Henrotin, A., Leclerc, S., Rezaei, N., et al. Lab-scale pilot for CO₂ capture vacuum pressure swing adsorption: Process simulations and global sensitivity analysis. Separation and Purification Technology, 2024(355): 124722. DOI: 10.1016/j.seppur.2024.124722.
[18] Gozálvez-Zafrilla, J.M., Bruni, L. and de Montigny, D. Uncertainty and global sensitivity analysis of a membrane biogas upgrading process. ChemEngineering, 2025,9(5):94. DOI: 10. 3390/chemengineering9050094.
[19] Zhao, Z., Li, Q., Shao, Y., et al. Prediction of inlet SO₂ concentration of wet flue gas desulfurization (WFGD) by operation parameters of coal-fired boiler. Environmental Science and Pollution Research, 2023(30):53089–53102. DOI: 10.1007/s11356-023-25988-5.
[20] Necula, S.C., Hauer, I., Fotache, D. and Hurbean, L. Advanced hybrid models for air pollution forecasting: Combining SARIMA and BiLSTM architectures. Electronics, 2025, 14(3): 549. DOI: 10. 3390/electronics14030549.
[21] Kurniawan, J.D., Parhusip, H.A. and Trihandaru, S. Predictive performance evaluation of ARIMA and hybrid ARIMA–LSTM models for particulate matter concentration. JOIN: Jurnal Online Informatika, 2024,9(2):259–268. DOI: 10.15575/join.v9i2.1318.