基于贝叶斯证据方法的裂解产物收率软测量模型

Soft-sensor models of pyrolysis product yields based on Bayesian Evidence framework

建立裂解产物收率的软测量模型对于乙烯裂解炉的生产具有重要的作用。采用支持向量回归方法可以建立较准确的乙烯裂解产品收率软测量模型,但是正则化参数等模型结构参数的选择对模型的精度仍然具有较大的影响。本文采用基于贝叶斯证据框架的支持向量回归方法,对正则化参数进行了迭代收敛计算,进而改进了乙烯裂解产品收率的软测量模型。在某工业乙烯裂解炉的生产数据上对产品收率进行了仿真验证,取得了良好的效果。

It is very important for ethylene pyrolysis furnace to build the soft-sensor models of product yields. These soft-sensor models can be developed based on support vector regression （SVR）, but the model accuracy is still influenced by the choice of hyper-parameters in the SVR. To address this problem, the support vector regression with Bayesian evidence framework is proposed here to optimize the regularization parameter itemtively. Thus the performance of the product yield soft-sensor models can be improved. The proposed method is illustrated on actual data set from an industrial pyrolysis furnace.