苯酚含量预测的高斯过程回归模型

Gaussian Process Regression Model for Prediction of the Content of Phenol

长期准确预测苯酚含量对双酚A生产过程的控制起着至关重要的作用。作为一种贝叶斯非参数模型,高斯过程本质上非常适合对长期持续的复杂过程进行建模。为此,提出一种基于高斯过程回归的苯酚含量预测模型。通过对高斯过程回归模型的协方差函数的选择与优化,在苯酚含量预测中取得了较好的测试结果。此外,采用ROC准则对生产过程的6个输入特征进行排序,并选择影响力较大的3个特征作为模型的输入变量,从而提高了模型的可解释性。

The long term precise prediction of the content of Phenol is essentially important for controlling the production process of Bisphenol-A （ BPA ）. As one kind of Bayesian nonparameteric model, Gaussian process is suitable for modeling of long term sustainable complex process. Thus, the prediction model of content of Phenol based on Gaussian process regression is proposed. Through the selection and optimization of covariance function of Gaussian process regression model, satisfactory test result is obtained in prediction of Phenol content. In addition, six of the input features of the production process are sequenced by adopting ROC criterion, and three of the features that offer greater influence are selected as the input variables of the model to enhance the interpretability of the model.