Abstract: The establishment of accurate BOF endpoint prediction model is particularly important for the improvement of production efficiency and liquid steel cleanliness. This paper took the converter of Shougang Jingtang duplex process in converter process as the research object, Pearson correlation analysis was conducted on historical production data, and 15 independent variables most relevant to the endpoint temperature and carbon content of the converter were obtained. Three machine learning algorithms of BP neural network, limit learning machine (ELM) and support vector machine (SVM) were used to build the prediction model of converter endpoint. Then 160 sets of new sample data were selected to verify the prediction accuracy of the three models. The results show that the accuracy of the prediction model of the end-point temperature and carbon content of the converter under the SVM model is higher. The hit rate of the prediction error of the endpoint temperature within ±15 ℃ is 90.6 %, and the hit rate of the prediction error of the endpoint carbon mass fraction within ±0.01 % is 93.8 %. In addition, the prediction model of converter end point based on support vector machine algorithm shows that the hit rate of endpoint temperature within ±15 ℃ and carbon mass fraction within ±0.01 % of duplex process in converter process is 9.1 % and 14.4 % higher than that of the conventional process.

Key words: duplex process in converter, correlation analysis, converter endpoint, machine learning, prediction accuracy