基于BT-SVM的氩氧精炼铬铁合金过程发生喷溅的预测方法

Splash Prediction Based on BT-SVM for Argon-Oxygen Decarburization Refining Cr-Fe Alloy Process

针对铬铁合金氩氧精炼过程中时常发生的喷溅现象,提出一种基于BT-SVM的喷溅预测方法。结合生产工艺,依据喷溅发生的原因及主要特征,选择了渣液上层表面温度与铁水温度等8个参数作为支持向量机的输入特征,选择爆发性喷溅、泡沫性喷溅、金属喷溅、正常工作作为输出特征,构建了基于BT-SVM的喷溅预测结构,将分类器分布在各个节点上,从而构成了多类分类支持向量机,并给出了分类函数的求解过程及其算法实现。测试结果表明:该方法可根据生产工艺参数实时预测喷溅是否发生,预测准确率在97%以上。

For the flash, a very common phenomenon in Cr-Fe alloy argon-oxygen decarburization refining process, here the splash-prediction method is proposed based on binary tree SVM （BT-SVM）. According to both the pro duction process and main features of splashing, eight parameters like surface temperature and the lower metal temperature of the slag liquid as input features for SVM were elected, and the explosive splash, bubble splash, metal splash and normal operation as the output features were chose. The splash-prediction structure based on BT- SVM, which puts the classifier on every node constitutes the multi class classification support vector machine, was established, correspondingly, both the algorithm and .solutions for classification function were given. The test results show that method can predict in real time if the slash will occur in views of the processing parameters, and the prediction accuracy rate is up to 97%. The method offers a good way to improve the product quality and ensure the work safety.