采用BP神经网络预测石脑油裂解烯烃收率

Prediction of olefin yield of naphtha steam cracking using back-propagation neural network

在乙烯裂解工业装置的典型操作条件下,分别选取正构烷烃、异构烷烃、环烷烃、芳烃为裂解原料,考察了这些模型化合物的蒸汽裂解产物分布情况。结果表明,正构烷烃是优质的乙烯裂解原料,乙烯收率为36%-45%;异构烷烃的丙烯收率约为23%,明显高于正构烷烃;环烷烃裂解乙烯和丙烯收率较低,丁二烯收率则较高,为14%-15%;芳烃很难裂解生成烯烃。建立了包含2个隐层的级联前向BP神经网络,以模型化合物和石脑油样本裂解烯烃收率为依据对该神经网络进行训练,确定了模型参数,并对2种石脑油的裂解烯烃收率仿真数据与实验结果进行了对比。结果表明,二者的误差小于1个百分点,该模型可用于预测石脑油裂解的烯烃收率。

The model compounds of normal paraffin,isoparaffin,cyclane and aromatics were selected as the feedstocks of steam cracking process and the distribution of the cracking products were investigated under the typical reaction conditions of ethylene cracking industial unit.The results showed that the normal paraffin was the quality feedstock for steam cracking,and its ethylene yield could reach 36%-45%.The propylene yield of isoparaffin was about 23%,which was higher than that of normal paraffin.The ethylene and propylene yields of cyclane were lower,but the butadiene yield was as high as 14%-15%.The aromatics were difficult to produce olefins by cracking.The back-propagation neural network containning two hiding layers were established and trained using the olefin yield of the model compounds and naphthas,and the parameters were determined.The simulation data and the experimental results of olefin yield for two naphthas were compared.The results indicated that the errors were less than 1 percent point.The model could be used to predict the olefin yield of naphtha cracking.