基于图像识别的铜转炉吹炼造渣期渣含Fe预测模型研究

铜转炉吹炼是火法炼铜的关键工序，其终点判断与炉寿、铜产率和直收率紧密相关，目前现有人工经验、仪器测定和物料平衡法等终点判断方法均存在一定的局限性。理论上铜转炉吹炼造渣期终点与渣含Fe是否达标有关，而不同Fe含量渣样呈现不同的图像特征，鉴于此，基于图形识别的特征向量提取原理，分别采用卷积神经网络(CNN)算法与支持向量机(SVM)算法，构建了铜转炉吹炼造渣期渣含Fe预测模型，为图像识别技术在铜转炉吹炼终点判断中的应用奠定数模基础。两种模型的实例分析表明，卷积神经网络的训练集预测准确率98%，测试集预测准确率约50%；支持向量机模型的训练集预测准确率99%，测试集预测准确率62%。

Research on Prediction Model of Fe Content in Slag during Copper Converter Slag-Making Period Based on Image Recognition

Copper converter blowing is the key process of pyrometallurgical copper smelting. Its end point judgment is closely related to furnace life, copper yield and direct yield. At present, the existing end point judgment methods such as manual experience, instrument measurement and material balance method have some limitations. Theoretically, the end point of copper converter slag blowing period is related to whether Fe content in the slag meets the standard, and the slag samples with different Fe content show different image features. In view of this, based on feature vector extraction principle of graphic recognition, the prediction model of Fe content in copper converter slag during slag blowing period is constructed by using convolution neural network (CNN) algorithm and support vector machine (SVM) algorithm respectively, It lays a digital and analog foundation for application of image recognition technology in judgment of blowing end point of copper converter. The instance analysis of two models shows that the prediction accuracy of training set of convolutional neural network is 98%, and the prediction accuracy of test set is about 50%; the prediction accuracy of training set of support vector machine model is 99%, and the prediction accuracy of test set is 62%.