一种基于Bagging算法的高斯过程集成建模方法

Ensemble modeling method based on Bagging algorithm and Gaussian process

针对单一高斯过程在化工过程软测量建模中存在估计精度不高的问题,利用Bagging和高斯过程回归算法,提出一种基于Bagging算法的集成高斯过程软测量建模方法.该算法使用Bagging技术从训练样本集中选取若干子训练样本集,利用该若干子集形成多个高斯过程模型,并通过加权组合方式进行集成,得到最终的模型输出.将该方法应用到某双酚A生产装置缩合反应釜出口24BPA含量的软测量建模中,仿真结果表明相比于单一高斯过程模型,该集成算法具有更高的精度和泛化能力.

In order to solve the problem of low estimation accuracy when using a single Gaussian process in the chemical process modeling, an ensemble model for soft-sensor is proposed based on Bagging and Gaussian process algorithms. The algorithm selects a number of sub-training sets from the whole training sample using Bagging technique, and then train the Gaussian process sub-models using the sub-training sets respectively. The output of the model is achieved by the integration with weighted outputs of the sub-models. Finally, the algorithm is applied to a soft sensor model of 24BPA in a bisphenol-A reactor. The simulation results show that the integration algorithm has higher accuracy and generalization ability comparing to a single Gaussian process model.