在基于聚类的多模型软测量建模中,如果数据聚类后类别之间界限不明显,常会出现样本类别误划分和类边界处样本估计不准确的问题。对此提出一种基于支持向量数据描述算法的聚类多模型软测量建模方法。该方法首先采用仿射传播聚类算法对数据进行类别划分,利用支持向量数据描述算法划定类边界从而确定样本与各类别的位置关系,根据样本与各类的位置关系信息从高斯过程回归算法建立的子模型、全局模型和局部模型中选择合适的模型对样本进行估计。将工业双酚A生产装置的现场数据建模和仿真,结果表明该方法是有效的。

In order to improve the accuracy and generalization ability of soft-sensor for complex industrial process, a Gaussian process ensemble soft-sensor modeling algorithm based on the improved bagging algorithm is proposed. This algorithm uses Gaussian process regression algorithm to build base learners and the resample method of bagging algorithm to form training subsets of base learners. A criteria for feature ordering base on normalized mutual information is proposed with selecting input features of base learners, which can implement supervised feature perturbance in the ensemble modeling for the sake of improving the diversity between base learners. When estimating the output of the test sample according to the output variances given by Gaussian process base learners, several base learners are selected adaptively to calculate the output of ensemble model. A soft-sensor modeling simulation using the data from the reactors of industrial Bisphenol-A production units shows the effectiveness of the algorithm.

For the multi-model soft-sensor modeling,its clustering results,sub-models modeling and combination of sub-models play an important role on the precision of soft-sensors. A multi-model soft-sensor modeling algorithm based on the improved affinity propagation clustering algorithm is proposed here. In order to improve the effect of clustering, an artificial fish-swarm algorithm is applied to optimize the preference parameter and the damping parameter in the affinity propagation clustering algorithm,and new overlapped clusters are built for the boundary samples located in the neighboring clusters. Then the support vector machine is used to build the regression sub-models for every cluster. The simulation results show that the algorithm for a standard data set and data of the industrial bisphenol-A production unit is effective.

为提高基于核函数的偏最小二乘算法非线性处理能力,削弱软测量模型对异常数据的敏感度,提高模型泛化能力,提出一种用于软测量在线建模的局部加权混合核偏最小二乘算法.该算法以多个具有不同特性的单一核函数构成混合核函数,将原始输入映射到高维特征空间,再采用局部加权学习算法在高维特征空间中计算样本权值,并对核变换后的样本数据进行加权处理,然后采用核函数偏最小二乘算法建立在线局部软测量模型.通过数值仿真和采用来自工业双酚A生产装置的现场数据进行在线软测量建模仿真,结果证明该算法是有效的.