| 引用本文: | 贺彦林,王晓,朱群雄.基于主成分分析--改进的极限学习机方法的精对苯二甲酸醋酸含量软测量[J].控制理论与应用,2015,32(1):80~85.[点击复制] |
| HE Yan-lin,WANG Xiao,ZHU Qun-xiong.Modeling of acetic acid content in purified terephthalic acid solvent column using principal component analysis based improved extreme learning machine[J].Control Theory and Technology,2015,32(1):80~85.[点击复制] |
|
| 基于主成分分析--改进的极限学习机方法的精对苯二甲酸醋酸含量软测量 |
| Modeling of acetic acid content in purified terephthalic acid solvent column using principal component analysis based improved extreme learning machine |
| 摘要点击 3430 全文点击 1805 投稿时间：2014-05-04 修订日期：2014-10-08 |
| 查看全文 查看/发表评论 下载PDF阅读器 |
| DOI编号 10.7641/CTA.2014.40398 |
| 2015,32(1):80-85 |
| 中文关键词 极限学习机 主成分分析 精对苯二甲酸 软测量 |
| 英文关键词 extreme learning machine principal component analysis purified terephthalic acid soft-sensing |
| 基金项目 家自然科学基金项目(61074153, 61473026)资助. |
|
| 中文摘要 |
| 目前, 化工生产过程日益复杂, 生产操作变量越来越多, 由于客观条件的限制, 有些重要的过程参数无法通过直接测量的手段精确测得. 通过软测量可实现复杂化工生产过程重要参数的精确测量, 进而指导化工企业的生产, 提高化工生产的产出效率, 是解决问题的一个有效的方法. 针对复杂化工过程软测量建模中存在的问题, 本文提出了一种改进的极限学习机模型(improved extreme learning machine, IELM). 一方面将主成分分析(principal component analysis, PCA)方法应用到极限学习机(ELM)里, 通过PCA对模型输入变量进行主成分分析, 不仅去除了变量间的线性相关关系, 而且对高数据进行降维处理, 最终降低了极限学习机的输入复杂性; 另一方面利用相关系数判断输入主元数据与输出数据间的相关关系, 从而得到正相关输入和负相关输入, 依据这两类数据构造ELM模型, 使得每类输入数据对网络的输出有同样的作用, 进一步提高极限学习机的泛化能力. 最后建立了PCA--IELM模型, 首先用标准数据库的Triazines数据集验证该模型有效性, 随后得出了基于PCA--IELM方法的精对苯二甲酸 (purified terephthalic acid, PTA)溶剂脱水塔塔顶醋酸含量软测量模型, 仿真结果表明PCA--IELM模型处理高维数据时较传统的ELM算法具有稳定性好, 建模精度高等特点, 为神经网络在复杂化工应用领域提供新思路. |
| 英文摘要 |
| Currently, chemical production processes are more and more complex, and there are more and more opration variables in chemical plants. Therefore, some important process variables can not be measured directly because of the limitations in practical conditions. A soft-sensing based method is adopted to realize that the production is measured accurately online, based on which the chemical company can enhance the production amount. To solve the problem, a principal component analysis (PCA) based improved extreme learning machine (IELM) soft-sensing model was proposed in this paper. On one hand, IELM method is combined with PCA to let the input values be analyzed by PCA for improving the generalization performance. On the other hand, the correlation coefficient was used to calculate the positive or negative relationship between the components and the outputs. Then the network strcture can be determined according to the poaitive components and the negative components. This structure has an advantage: the input components have the same effection on the outputs, which can enhance the performance of ELM. Finally, PCA--IELM model was built, and the Triazines dataset from UCI standard database was selected to verify the effectiveness of this model. Then the PCA--IELM was used as a soft sensor for modeling purified terephthalic acid (PTA) solvent column acetic acid content. The experimental results show that when dealing with high-dimensional data the PCA--IELM has better and higher precision modeling than the ELM. The PCA--IELM model provides a new idea for neural networks applying to complex chemical processes. |
|
|
|
|
|