一种基于KPCA-BN的软测量建模方法

A soft sensing method base on KPCA and BN

本文提出了一种将核主元分析(KPCA)和贝叶斯网络(BN)相结合的软测量建模方法。核主元分析可对样本数据进行特征提取,消除数据之间的相关性,降低网络模型的输入变量维数。然后利用贝叶斯网络进行建模,采用基于剪枝算法的EM算法求解高斯混合模型的参数,再利用高斯混合模型逼近贝叶斯网络中变量的联合概率密度,训练贝叶斯网络,该方法不仅降低了模型的复杂性,而且提高了模型的泛化能力。最后采用该方法建立乙烯精馏塔中乙烷浓度的软测量模型,结果表明基于KPCA-BN方法建立的软测量模型有更好的预测效果和泛化能力,是一种有效的数据建模方法。

A kind of soft sensing modeling method which is combination of kernel principal component analysis （KPCA） with bayesian network （BN） is proposed. The KPCA is used to make the feature extraction, eliminate the correlation of the input sample data and decrease the dimensions of input variable in the network model. Then the BN is applied to build a model. After determining the parameters of the gaussian mixture model by EM algorithm based on the pruning algorithm, the gaussian mixture model is used to approximate the probability density of the variables in the bayesian network and train the bayesian network, which can not only reduce the complexity of model but also improve the generalization ability. Finally the proposed method is used to build soft sensing model of the concentration of ethane in the ethylene distillation, and the results show that KPCA-BN approach has a better prediction and generalization ability, and it is an effective data modeling approach.