Comprehensive status evaluation and prediction of blast furnace based on cascade system and combined model

The comprehensive status of blast furnaces was one of the most important factors affecting their economy, quality, and longevity. The blast furnace comprehensive status had the nature of “black box,” and it was “unpredictable.” In this study, a blast furnace comprehensive status score and prediction method based on a cascade system and a combined model were proposed to address this issue. A dual cascade evaluation system was developed by integrating subjective and objective weighting methods. The analytic hierarchy process, coefficient of variation, entropy weight method, and impartial combinatorial weighting were jointly employed to determine the optimal weight distribution across indicators. Categorized statuses (raw material, gas flow, furnace body, furnace cylinder, and iron slag) were evaluated. Based on the five categories of the status data, the second cascade was applied to upgrade the quantitative evaluation of the comprehensive status. The weights of the different categories were 0.22, 0.15, 0.22, 0.21, and 0.20, respectively. According to the data analysis, the results of the comprehensive status score closely matched the on-site production logs. Based on the blast furnace smelting period, the maximal information coefficient method was applied to the 100 parameters that were most relevant to the comprehensive status. A combined prediction model for a comprehensive status score was designed using bidirectional long short-term memory (BiLSTM) and categorical boosting (CatBoost). The test results indicated that the combined model reduced the mean absolute error by an average of 0.275 and increased the hit rate by an average of 5.65 percentage points compared to BiLSTM or CatBoost alone. When the error range was ±2.5, the combined model predicted a hit rate of 91.66% for the next hour’s comprehensive status score, and its high accuracy was deemed satisfactory for the field. SHapley Additive exPlanations and regression fitting were applied to analyze the linear quantitative relationship between the key variables and the comprehensive status score. When the furnace bottom center temperature was increased by 10°C, the comprehensive status score increased by 0.44. This method contributes to a more precise management and control of the comprehensive status of the blast furnace on-site.