Abstract:  In order to improve the end point temperature control level of molten steel in LF refining， a combined method based on principal component analysis （ PCA ） and BP neural network was proposed to predict the end-point temperature of molten steel in LF ladle furnace.Based on the metallurgical theory and practical production practices， 10 factors that have significant influence on the endpoint temperature of 42CrMo steel production process were selected as the index system of the prediction model.Then the data were processed by principle component analysis， and seven principal component variables were obtained. The cumulative variance contribution rate was 87.24 %， and the correlation between the data was eliminated.Based on this， a prediction model of end point temperature of LF furnace based on PCA-BP neural network was established. When the prediction error of the model is within ± 25 °C， the hit rate of the model is 98.71 %. The model has good recognition ability and can achieve the purpose of predicting the end point temperature of LF furnace production process.

Key words: Principal Component Analysis, BP Neural Network, Endpoint Temperature Prediction

摘要： 为提高LF精炼钢水终点温度控制水平，提出了基于主成分分析（PCA）和BP神经网络的联合方法预测LF钢包炉精炼钢水终点温度。基于冶金理论和实际生产实践，选取了42CrMo钢生产过程的10个对终点温度有显著影响的因素作为预测模型的指标体系，然后借助主成分分析法对样本数据进行处理，得到了7个主成分变量，累计方差贡献率为87.24%，消除了数据之间的关联性，以此为基础，建立了基于PCA-BP神经网络的LF炉终点温度预测模型，该模型预测误差在±25 ℃时，模型的命中率为98.71%，模型有较好的识别能力，能够达到LF炉生产过程预测终点温度的目的。

关键词: 主成分分析, BP神经网络, 终点温度预测