催化重整装置产品质量软测量技术应用

Soft-sensing Technology for Product Quality Forecasting in Catalytic Reforming Process

采用机理方法、线性拟合方法和一阶TSK模糊神经网络算法分别对催化重整装置的主要产品质量指标——芳烃收率建立软测量模型,并在国内某大型工业级催化重整装置上进行在线应用。研究结果表明:进料负荷变化不大时3,种模型的在线预测趋势均能较好地跟踪芳烃收率的实际变化;当进料负荷变化较大时,3种模型的预测偏差分别为0.24wt%、0.67wt%和0.95wt%,且只有机理模型能如实地反映芳烃收率的变化。这说明机理模型具有预测精度高及外推性能好等优势,而且机理模型对样本数目要求最少,其预测精度基本不受样本数的影响。

Basing on mechanism method,linear regression method and first-order TSK fuzzy neural network algorithm,three soft-sensing models for aromatics yield in catalytic reforming process were presented respectively.Their domestic application in a large-scale catalytic reforming process shows that at little change in the feed charge,three models can track the actual change of aromatics yield;and at big change in the feed charge,their average deviations stand at 0.24wt%,0.67wt% and 0.95wt% respectively,and only mechanism model can reflect the actual change of aromatics yield.This means that mechanism model which asking for least samples can outperform others in forecast precision.