Prediction of SOx–NOx emission from a coal-fired CFB power plant with machine learning: Plant data learned by deep neural network and least square support vector machine. (10th October 2020)

• Record Type: Journal Article
• Authors: Adams, Derrick; Oh, Dong-Hoon; Kim, Dong-Won; Lee, Chang-Ha; Oh, Min
• Is Part Of: Journal of cleaner production. Volume 270(2020)
• Publication Date: 2020-10-10
• Subjects: SOx-NOx emissions -- Deep neural network -- Least-square support vector machine -- Coal-fired circulating fluidized bed boiler -- Energy production -- Data featuring; Factory and trade waste -- Management -- Periodicals; Manufactures -- Environmental aspects -- Periodicals; Déchets industriels -- Gestion -- Périodiques; Usines -- Aspect de l'environnement -- Périodiques; 628.5
• DOI: 10.1016/j.jclepro.2020.122310
• Languages: English
• ISSNs: 0959-6526

Abstract

Abstract: The circulating fluidized bed boiler is an advanced clean energy technology that has received much attention in the power industry due to its fuel flexibility. In this study, a deep neural network with a modified early stopping algorithm and least square support vector machine were developed to predict SOx and NOx emissions associated with coal conversion in energy production. The models were trained on commercial plant data and the effect of dynamic coal and limestone properties (which is assumed constant in the literature) such as proximate analysis, ultimate analysis and particle size distribution on prediction accuracy were investigated. The results revealed that training the models without the assumptions improved the accuracy of the testing phase by at least 10% and 40% with a coefficient of efficiency of 0.8925 and 0.9904 for SOx and NOx respectively. In addition, interactive and pairwise correlation featuring were implemented which gave a maximum computational time reduction of 46.67% for NOx emission prediction. The developed models and findings can be applied not only for online operation and optimization of a coal-fired CFB boiler with high accuracy but also in the scale-up of power production at a low computational cost. Highlights: Developed DNN and LSSVM to predict SOx-NOx emissions accompanying coal conversion. The effect of coal properties on prediction accuracy of models is investigated. Accuracy improvement and computational time reduction strategies are conducted. The models achieved 46.67% maximum computational time reduction with filtering. The developed models can successfully augment CEMS for SOx–NOx monitoring.