中文摘要：

针对目前软测量建模过程中,单个模型难以精确描述复杂非线性对象而多模型又多采用静态模型因而对系统实际运行中的动态变化考虑不足的问题,提出了一种基于多模型动态融合的自适应软测量建模方法。该方法首先采用仿射传播聚类算法对样本数据进行分类,并对不同类别的输入样本分别建立基于高斯过程回归的子模型,最后使用动态Gauss-Markov估计对各子模型估计值进行融合。将上述方法应用于对二甲苯（p-xylene,简称PX）吸附分离过程纯度的软测量建模,仿真结果表明该方法能够有效地增强模型适应工况变化的能力,是一种有效的软测量建模方法。

英文摘要：

Single-model soft sensors in soft sensor modeling are difficult to accurately describe complex nonlinear objects, while multi-model soft sensors usually use several static models, which cannot reflect the dynamic characteristics of industrial processes. An adaptive soft sensor model based on the multi-model dynamic fusion method was built which was implemented by an adaptive Gauss-Markov estimation method proposed in this study. The input samples of the model were clustered by affinity propagation algorithm. Sub-models were built for each clustering based on Gaussian process regression algorithm. Moreover, the model outputs were predicted by fusing the values of sub-models dynamically based on the adaptive Gauss-Markov estimation. This soft sensor model was applied to predict p-xylene（PX） purity in adsorption separation processes. The results indicate that the proposed model-building method actually increased the adaptive ability of the model under various operation conditions.