Evidence reasoning-based alarm method for silicon content in blast furnace hot metal using unbalanced monitoring samples

Abstract

Silicon content monitoring and alarm based on data-driven methods is an essential approach to ensure product quality and smelting status in the blast furnace ironmaking process. However, due to the frequent fluctuation of production process, the sample data of silicon content usually shows obvious imbalance characteristics, i.e., less abnormal samples and more normal samples, posing great challenge to alarm task. In this case, this paper develops a novel data-driven alarm method based on the evidence reasoning (ER) rule. First, a normalization method of likelihood probability is given to construct the offline referential evidence matrix (REM) for characterizing the imbalanced sample distributions. Second, the online sample matches with the REM to generate the corresponding alarm evidence, and then the iterative ER fusion strategy is proposed to combine the historical and current evidence. Finally, the alarm decision can be made according to the fusion results. The integration of current and historical information helps to reduce the false alarms/missed alarms caused by the imbalanced samples and frequent state switches. The proposed ER rule-based monitoring approach is tested on real industrial data and comparison results with conventional methods like filtering and delay timer show that it has better comprehensive performance.