A blast furnace coke ratio prediction model based on fuzzy cluster and grid search optimized support vector regression

In the study of blast furnace coke ratio, existing methods can only predict coke ratio of daily. At the same time, the data under abnormal furnace conditions are excluded, and the model’s robustness needs to be improved. In order to improve the prediction accuracy and time precision of the blast furnace mathematical simulation model, a blast furnace coke ratio prediction model based on fuzzy C-means (FCM) clustering and grid search optimization support vector regression (SVR) is proposed to achieve accurate prediction of coke ratio. First, preprocess the blast furnace sensor data and steel plant production data. Then, the FCM algorithm is used to cluster the data under different furnace conditions. Finally, the SVR model optimized by grid search is used to predict the coke ratio under different blast furnace conditions. The average absolute error of the improved model is 1.7721 kg/t, the hit rate within 0.5% error is 81.19%, the coefficient of determination R2 is 0.9474, and the prediction performance is better than ridge regression and decision tree regression. Experiments show that the model can predict the coke ratio of molten iron in each batch when the blast furnace conditions are going forward and fluctuating, and it has high time accuracy and stability. It objectively describes the changing trend of blast furnace conditions, and provides new research ideas for the practical application of blast furnace mathematical models.