Abstract: The converter smelting process contains complex multi-phase and high-temperature physical and chemical reac tions， and it is of great significance to establish a reliable converter endpoint prediction model to effectively reduce the fluc tuation of molten steel composition and improve the quality of steel. Based on the actual production data of a 200 t con⁃ verter in a steel mill， the particle swarm optimization algorithm is used to select the optimal penalty parameter C and ker nel parameter g of the support vector machine model to establish a prediction model， and the carbon mass fraction and tem perature at the end point of the converter are predicted. After data processing 425 sets of data were obtained and divided into training set data and test set data， and normalized them， of which 50 groups were randomly selected as test set data. The results show that the accuracy of carbon mass fraction （error ±0. 015%） and temperature （error ±15 ℃） is 81. 8% and 80% respectively. Compared with BP neural network model and RBF model， support vector machine model optimized by particle swarm optimization has higher accuracy and better generalization ability.

Key words: Converter Steelmaking, PSO-SVM Model, Endpoint Temperature, End-point Carbon Content, Predictive Models

摘要： 转炉冶炼过程包含着复杂的多相、高温的物理化学反应，建立可靠的转炉终点预测模型对有效减少钢水成 分波动、提高钢铁品质有重要的意义。以某钢厂200 t转炉实际生产数据为依据，采用粒子群优化算法选取支持向 量机模型最优惩罚参数C和核参数g的方法建立预测模型，对转炉终点碳质量分数和温度进行预测。将数据处理 后得到425组数据，数据划分为训练集数据和测试集数据，并对其进行归一化预处理，其中，随机选取50组为测试 集数据。结果表明，转炉终点预测模型的终点钢水碳含量（误差±0. 015%）的命中率为84%，终点温度（误差±15 ℃） 的命中率为80%。与BP神经网络模型和RBF模型相比，基于粒子群算法优化的支持向量机模型具有精度高、泛化 能力强的特点。

关键词: 转炉炼钢, PSO-SVM模型, 终点温度, 终点钢水碳含量, 预测模型