Iron-making Process Monitoring based on Wasserstein Dynamic Stationary Subspace Analysis

The detection of early abnormalities in blast furnaces is important for the safety of iron-making processes, the reduction of energy consumption, and the improvement of economic benefits. Due to the fluctuation of raw material composition, production schedule adjustment, equipment degradation, etc., the iron-making process data often present non-stationary characteristics, resulting in the time-varying statistical characteristics of the data even under normal conditions. To solve this problem, a dynamic stationary subspace analysis method based on Wasserstein distance to detect anomalies of the iron-making process is proposed. Firstly, the switching time of hot blast stoves is identified to reduce the false alarms caused by the hot blast stove switching; Secondly, the differences in data distribution of different time intervals are measured by the Wasserstein distance, and the stationary features are obtained by minimizing the Wasserstein distance between intervals; Finally, the abnormal furnace condition monitoring index based on Mahalanobis distance is constructed by using the stationary features. The simulation results based on an actual iron-making process dataset show that the method proposed in this paper can effectively monitor the abnormal furnace conditions.