基于灰色关联分析和机器学习的高炉铁水硅含量预测

Prediction of Silicon Content in Blast Furnace Molten Iron Using Grey Correlation Analysis and Machine Learning

高炉铁水硅含量通常被用作表征“炉温”和高炉经济效益的重要指标,建立及时准确的高炉铁水硅含量预报模型,对于高炉稳定生产具有重要意义。本工作考虑时间延迟及灰色关联分析(GRA)并结合主成分分析(PCA)和长短期记忆网络(LSTM)建立了一种新的预测模型。首先,从硅的还原氧化行为角度出发,选出高炉生产中与铁水硅含量相关的操作参数、状态参数和结果参数。其次,通过GRA确定各参数与铁水硅含量的关联度大小,筛选出硅含量的主要影响因素。在此基础上,对国内某3200 m^(3)级高炉一年共计8544组生产数据进行降维处理。模型输入变量由降维得到的13个主成分组成,以铁水硅含量作为输出变量,建立了预测模型。该预测模型具有合理的泛化能力、鲁棒性和准确性。预测结果分析表明,该模型的决定系数(R^(2))为0.9297,均方误差(MSE)和平均绝对误差(MAE)分别为0.0012和0.0254,实现了对高炉铁水硅含量精确预测。

The silicon content in molten iron is commonly used as an important indicator to characterize the‘chemical heat’and economic benefits of the blast furnace.Establishment a timely and accurate prediction model for the silicon content in molten iron is of great significance for stable production of the blast furnace.In this work,an innovative prediction model was established by considering the time delay and the grey relational analysis(GRA)combining principal component analysis(PCA)and long short-term memory network(LSTM).Firstly,from the perspective of the reduction-oxidation behavior of silicon,the operating parameters,state parameters and result parameters relevant to the silicon content of blast furnace iron in production were selected.Secondly,through GRA,the degree of correlation between each parameter and the silicon content in molten iron was determined,and the main factors affecting the silicon content in molten iron were screened out.On this basis,a total of 8544 sets of production data for a 3200 m^(3)blast furnace in China were dimensionally reduced in one year.A prediction model was established using the 13 principal components obtained as input variables and the silicon content in molten iron as output variables.The prediction model has reasonable generalization ability,robustness and accuracy.The analysis of prediction results shows that the determination coefficient(R^(2))of the model is 0.9297,the mean square error(MSE)and mean absolute error(MAE)are 0.0012 and 0.0254 respectively,realizing accurate prediction of the silicon content of blast furnace molten iron.