Prediction of Blast Furnace Fuel Ratio Based on Back‐Propagation Neural Network and K‐Nearest Neighbor Algorithm. Issue 10 (28th July 2022)

• Record Type: Journal Article
• Authors: Zhang, Longyao; Jiao, Kexin; Zhang, Lei; Zhang, Jianliang; Sun, Minmin; Zhou, Zhenhao; Zheng, Anyang
• Is Part Of: Steel research international. Volume 93:Issue 10(2022)
• Page Start: n/a
• Page End: n/a
• Publication Date: 2022-07-28
• Subjects: back propagation neural networks -- blast furnaces -- fuel ratio predictions -- k -nearest neighbors; Steel -- Periodicals; Steel -- Metallurgy -- Periodicals; 669.142
• DOI: 10.1002/srin.202200215
• Languages: English
• ISSNs: 1611-3683

Abstract

Abstract : In recent years, CO2 emissions from the industry have gradually become a major concern, with CO2 from blast furnace production processes being an important source. If the key factors affecting the fuel ratio in the production process can be found and predicted, it can effectively guide the blast furnace production and reduce CO2 emissions. Herein, 55 production parameters that are important in the operation of a blast furnace are analyzed, and high‐correlation filtering and reverse feature elimination to find the four parameters that have the most influence on the fuel ratio of the blast furnace are used. These parameters are oxygen enrichment rate, raceway flame temperature, per ton of iron blast consumption, and coke ratio. Finally, two models, back propagation (BP) neural network and k ‐nearest neighbor are used to predict the fuel ratio using these four parameters. Using BP is more effective than using K‐nearest neighbor (KNN); the accuracy of BP is 93.02%, and KNN is 90.48% when the error is within 2%. Abstract : Correlations are analyzed for 55 key parameters in blast furnace operation, and reverse feature elimination is performed. Four of these parameters are selected as having the most important impact on the fuel ratio. The fuel ratio is predicted using a back propagation neural network as well as k ‐nearest neighbor regression, and the accuracy can reach 93.02%.