基于火焰特征提取和PSO-ELM算法的转炉碳温预测  

Prediction model of converter temperature and carbon content based on spectral feature extraction and PSO-ELM

作　　者：张强 杨勇[1] 戴雨翔 赵博[1] 林路[1] ZHANG Qiang;YANG Yong;DAI Yuxiang;ZHAO Bo;LIN Lu(Metallurgical Technology Institute,Central Iron and Steel Research Institute Co.,Ltd.,Beijing 100081,China)

机构地区：[1]钢铁研究总院有限公司冶金工艺研究所,北京100081

出　　处：《钢铁研究学报》2025年第2期141-150,共10页Journal of Iron and Steel Research

基　　金：国家自然科学基金资助项目(51874102,52074093,52274329)。

摘　　要：转炉炉口火焰的颜色和纹理与转炉碳含量及温度存在密切的联系,通过火焰特征提取和神经网络模型预测转炉碳温为转炉终点控制提供了新的思路。基于转炉炉口火焰光谱数据集以及PSO-ELM神经网络建立了转炉碳含量及温度预测模型。针对原始光谱中含有较多的噪声、杂散光等问题,采用小波算法对光谱数据集进行降噪处理。由于炉口火焰光谱数据量大,冗余信息较多,采用Skowron差别矩阵的属性约简算法从给定的2048维波长数据中找到对决策结果具有显著影响的8个特征指标。通过计算8个特征指标的MIC系数,证明所选指标具有独立性与非共线性,避免了因为指标之间高度相关导致建模不稳定以及过拟合的风险。基于PSO-ELM算法建立了预测模型,针对ELM在初始化时随机产生输入权值和隐含层阈值的缺陷,采用粒子群算法进行了优化。通过将PSO-ELM模型应用到转炉碳温的预测中,实例验证表明该模型在碳温预测上的精度较高,预测效果良好,适用于转炉碳温预测,有较好的工程应用前景。The color and texture of the flame at the converter mouth are closely related to the carbon content and temperature of the converter.The prediction of the carbon content and temperature of the converter through the flame characteristics of the converter mouth collected by the spectrometer provides a new idea for the end point control of converter steelmaking.Based on the flame spectrum data set of the converter mouth and the PSO-ELM neural network,a prediction model of the carbon content and temperature of the converter is established.In view of the fact that the original spectrum contains more noise,stray light,etc.,wavelet algorithm is used to reduce the noise of the spectral data set.Due to the large amount of flame spectrum data at the converter mouth and the large amount of redundant information,the attribute reduction algorithm of the Skowron difference matrix is used to find the smallest data set with the smallest decision set coverage from the given 2048-dimensional wavelength data,and 8 special diagnosis indicators are obtained.By calculating the MIC coefficients of the 8 indicators,it is proved that the selected indicators are independent and non-collinear,avoiding the risk of unstable modeling and overfitting due to the high correlation among the indicators.A prediction model is established based on the PSO-ELM neural network,and the particle swarm optimization algorithm is used to optimize the defects of the input weights and hidden layer thresholds randomly generated by the ELM during initialization.By applying the PSO-ELM model to the prediction of carbon temperature of converter,the example validation shows that the model has high accuracy and good prediction effect on carbon temperature prediction,which is suitable for the prediction of carbon temperature of converter and has a better engineering application prospect.

关 键 词：转炉冶炼 终点控制 机理模型 炉口火焰识别 ELM神经网络 

分 类 号：TF713[冶金工程—钢铁冶金]