Abstract: In views of the normalizing annealing furnace roll marks problem occurred in the process of normalizing annealing of silicon steel in Baosteel, by combining big data from actual production with product quality problems, data mining and data analysis methods were applied to actual production to solve the pain points and provide decision support, a robust and practical data analysis method for the steel production process has been developed, which avoided the previous problems of omission and nonquantitative judgment through manual identification. Through the intelligent decision system platform to obtain the actual production and surface detector data, variable selection and feature engineering were carried out based on Pearson correlation coefficient algorithm and applied random forest algorithm to classify and forecast the data, which realized the traceability and monitoring of quality problems. Therefore the quality problem whether the defect of furnace roll marks can be eliminated by rolling would be predicted by quantitative data, and the recognition accuracy reached 96.43 %.

Key words: silicon steel, roll marks, data analyzing, data mining