| 引用本文: | 胡磊,刘强,吴永建,范自柱.间歇过程动态潜结构阶段划分与在线监控[J].控制理论与应用,2022,39(2):307~316.[点击复制] |
| HU Lei,LIU Qiang,WU Yong-jian,FAN Zi-zhu.Dynamic latent structure based phase partition and online monitoring for batch processes[J].Control Theory and Technology,2022,39(2):307~316.[点击复制] |
|
| 间歇过程动态潜结构阶段划分与在线监控 |
| Dynamic latent structure based phase partition and online monitoring for batch processes |
| 摘要点击 1456 全文点击 386 投稿时间：2020-11-18 修订日期：2021-04-02 |
| 查看全文 查看/发表评论 下载PDF阅读器 |
| DOI编号 10.7641/CTA.2021.00817 |
| 2022,39(2):307-316 |
| 中文关键词 间歇过程 过程监控 动态主元分析 相似性度量 |
| 英文关键词 batch processes process monitoring dynamic principal component analysis similarity measure |
| 基金项目 国家重点研发计划(2020YFB1710003), 国家自然科学基金项目(U20A20189, 61991401, 62161160338, 61673095), 兴辽英才计划(XLYC1907049, XLYC18080), 教育部基本科研业务费项目(N180802004)资助. |
|
| 中文摘要 |
| 阶段划分是间歇过程准确建模和有效监控的前提. 针对传统阶段划分方法未考虑间歇过程的动态性造成
阶段划分不准确、影响监控精度, 且具有参数选择难、鲁棒性差的局限, 提出一种基于动态潜结构的动态间歇过程
阶段划分与在线监控方法. 首先, 对间歇过程三维张量数据沿变量方向展开, 并增加时滞变量构建增广矩阵来提取
过程动态关系; 然后, 以增广矩阵作为输入, 定义一种新的基于解释方差变化的合并代价函数, 衡量不同子序列之间
的动态潜结构相似度; 利用上述动态潜结构相似度的衡量标准, 提出基于自底向上启发式搜索策略的动态间歇过
程阶段划分方法; 最后, 对划分得到的各阶段分别建立基于动态主元分析的子阶段模型和统计指标来实现在线监
控. 采用青霉素补料分批发酵过程数据开展实验研究, 结果表明了所提方法的有效性和优越性. |
| 英文摘要 |
| The phase partition is essential for statistical process modeling and monitoring of multi-phase batch processes.
Traditional methods do not take into account the dynamics that make the phase partition and monitoring be inappropriate.
Also it is difficult to determine model parameters and obtain fair performance when the underlying data structure is complex.
A dynamic latent structure based phase partition and online monitoring method is thus proposed for dynamic batch
processes. First, three-dimensional data are unfolded in variable-wise direction and original variables are augmented in
order to extract the dynamic relations. After that, a new cost function based on relative variations of explained variance is
defined in this paper to measure the difference of dynamic latent structure between sub-sequences. Using the defined cost
function, a bottom-up heuristic algorithm is employed to find out the sub-optimal stage division. Finally, dynamic principal
component analysis based sub-models and statistical indices for each phase are derived for online monitoring of dynamic
batch processes. The application results on penicillin batch fermentation demonstrate the effectiveness and superiority of
the proposed method. |
|
|
|
|
|