摘要
针对传统聚类方法过于依赖数据空间分布和先验知识的缺点,该文提出了一种改进的扩张搜索聚类算法。该算法充分考虑了样本疏密度对聚类效果的影响,根据每个样本点不同的疏密度给予不同的搜索半径,并且引入阈值对不同疏密度的样本点采用不同的聚类方式,以适于各种形状的样本分布。采用这种改进的扩张搜索聚类算法对样本数据进行聚类,并用高斯过程回归(GPR)对各类样本子集分别建立对应的软测量子模型,最后采用开关切换的多模型融合方式得到最终的软测量多模型。以双酚A生产过程结晶单元的仿真结果为例,对装置出口处的苯酚浓度进行软测量建模,获得了较好的实验结果。
An improved expanding searching clustering algorithm is proposed to overcome the shortcomings of the traditional clustering methods relying on data space distribution and prior knowledge too much. In consideration of the effects of the sample density on the searching radius,the improved algorithm selects different searching radius according to the density of each sample point.For all sample distribution shapes,the threshold value is applied to choose different clustering methods relying on different density points. Sample data is clustered by using the improved expanding searching clustering algorithm. All soft sensor models are built up by Gaussian process regression (GPR) . The final model is formed by using the switch fusion mode according to the results of clustering. A sample of a bisphenol-A production crystallization unit is applied to make a simulation for building the soft-sensor model of the phenol concentration at the exit device and the good experiment results are obtained.
出处
《南京理工大学学报》
EI
CAS
CSCD
北大核心
2017年第5期574-580,共7页
Journal of Nanjing University of Science and Technology
基金
国家自然科学基金(61273070)
江苏省高校优势学科建设工程资助项目
关键词
疏密度
阈值
高斯过程回归
扩张搜索聚类算法
软测量建模
densities
threshold
Gaussian process regression
expanding searching dusting algorithm
soft sensor modeling