选择性集成LTDGPR模型的自适应软测量建模方法
发布时间：2018-11-16 16:25
【摘要】：随着时间的增加,传统时间差(TD)模型会出现性能显著下降的问题。为了提高TD模型的可靠性和预测精度,同时考虑过程的时滞特征,基于一种选择性集成策略,提出一种局部时间差高斯过程回归(LTDGPR)模型的自适应软测量建模方法。首先,提取出数据库中的时滞动态信息,对建模数据进行重构;然后,采取局部化策略对差分后的重构样本进行统计划分,得到LTDGPR模型集。对于新来的输入样本,选择部分泛化能力强的LTDGPR模型进行集成,估计出含一定时间差的主导变量动态偏移值;最后,基于TD模型思想对当前时刻主导变量值进行在线预测。通过脱丁烷塔过程的数据建模仿真研究,验证了所提方法的有效性和精度。
[Abstract]:With the increase of time, the performance of the traditional time-difference (TD) model will decrease significantly. In order to improve the reliability and prediction accuracy of TD model and take into account the time-delay characteristics of the process, an adaptive soft-sensor modeling method for Gao Si regression (LTDGPR) model with local time difference is proposed based on a selective integration strategy. First, the time-delay dynamic information is extracted from the database, and then the modeling data is reconstructed, and then the LTDGPR model set is obtained by statistical partitioning of the reconstructed samples after the difference by using the localization strategy. For the new input samples, the LTDGPR model with strong generalization ability is selected for integration, and the dynamic offset value of the dominant variable with certain time difference is estimated. Finally, based on the idea of TD model, the dominant variable value at the current time is predicted online. The validity and accuracy of the proposed method are verified by the data modeling and simulation of the debutane column process.
【作者单位】： 江南大学物联网工程学院自动化研究所;轻工过程先进控制教育部重点实验室;
【基金】：国家自然科学基金项目(21206053,21276111) 江苏省“六大人才高峰”项目(2013-DZXX-043)~~
【分类号】：TQ018
，
本文编号：2336011
[Abstract]:With the increase of time, the performance of the traditional time-difference (TD) model will decrease significantly. In order to improve the reliability and prediction accuracy of TD model and take into account the time-delay characteristics of the process, an adaptive soft-sensor modeling method for Gao Si regression (LTDGPR) model with local time difference is proposed based on a selective integration strategy. First, the time-delay dynamic information is extracted from the database, and then the modeling data is reconstructed, and then the LTDGPR model set is obtained by statistical partitioning of the reconstructed samples after the difference by using the localization strategy. For the new input samples, the LTDGPR model with strong generalization ability is selected for integration, and the dynamic offset value of the dominant variable with certain time difference is estimated. Finally, based on the idea of TD model, the dominant variable value at the current time is predicted online. The validity and accuracy of the proposed method are verified by the data modeling and simulation of the debutane column process.
【作者单位】： 江南大学物联网工程学院自动化研究所;轻工过程先进控制教育部重点实验室;
【基金】：国家自然科学基金项目(21206053,21276111) 江苏省“六大人才高峰”项目(2013-DZXX-043)~~
【分类号】：TQ018
，
本文编号：2336011
本文链接：https://www.wllwen.com/kejilunwen/huaxuehuagong/2336011.html
