Evidence fusion of process variable and expert knowledge for abnormal condition recognition in blast furnace ironmaking process

Accurately recognising blast furnace condition is essential for effectively monitoring the status of the blast furnace ironmaking process. However, numerous process variables and complex mechanisms of blast furnace ironmaking process make the recognition tasks challenging. Therefore, this article proposes a novel abnormal condition recognition method by fusing process variables with expert knowledge through evidence fusion. Firstly, an objective evidence model is developed by capitalising process variables, using a support vector classifier with tunable parameters and a Platt scaling probability mapping function. Concurrently, a subjective evidence model is constructed by leveraging expert knowledge, incorporating multiple logical rules and membership functions designed for specific blast furnace condition. To address potential conflicts between the evidences produced by the above objective and subjective evidence models, a conflict redistribution and evidence fusion method is invented. The resulting fusion of subjective and objective evidences facilitates accurate blast furnace condition recognition. Industrial experimental results validate the effectiveness of the proposed method in recognising abnormal conditions, such as hanging and channelling.