Jue Tang, Si-nan Li, Quan Shi, Zhen Zhang, Yue-song Qi, Man-sheng Chu, Hong-yu Tian
Blast furnace (BF) operation state was difficult to characterize, measure, and predict. To solve this problem, an intelligent evaluation and advanced prediction method of BF operation state based on industry big data and machine learning was proposed. Based on the criteria of high productivity, low consumption, high quality, smooth running and long life, five BF parameters were extracted according to production experience and metallurgy process. Using the unsupervised learning, a 4-grade evaluation rule was established to realize the intelligent rating of BF operation state. Based on Kendall and maximal information coefficient, 70 BF parameters with the most characteristic power of BF operation state were determined. The weights of BF parameters were calculated by applying the criteria importance through intercriteria correlation and the grey correlation degree. The weights of raw material, fuel, gas distribution, cooling stave, BF hearth, and iron and slag were 0.241, 0.213, 0.140, 0.098, 0.117 and 0.191, respectively. The weight of data interval was calculated by using the grading algorithm and the monotonicity, and then, the intelligent scoring mechanism based on the multiple weights was formed. It was beneficial to qualitatively and quantitatively characterizing the “black box” BF operation state. Furthermore, combining the algorithm and the evaluation mechanism, a graded prediction model of BF operation state was developed and proposed. It was shown that, compared with the conventional prediction model, the mean absolute error and mean square error of the graded prediction model were reduced by 0.35 and 1.29, respectively, while the explained variation was increased by 14.56%, the hit rate was increased by 5.1% within the error of 3%, and the average hit rate was more than 90.6%. It could be applied to reliably predict the score of BF operation state in the next hour and accurately provide the support for the practical controlling of the running BF.