| 深度融合特征提取网络及其在化工过程软测量中的应用 |
| |
| 引用本文： | 周乐,沈程凯,吴超,侯北平,宋执环.深度融合特征提取网络及其在化工过程软测量中的应用[J].化工学报,2022,73(7):3156-3165. |
| |
| 作者姓名： | 周乐 沈程凯 吴超 侯北平 宋执环 |
| |
| 作者单位： | 1.浙江科技学院自动化与电气工程学院，浙江 杭州 310024;2.浙江大学工业控制技术国家重点实验室，浙江 杭州 310027 |
| |
| 基金项目： | 国家自然科学基金项目(62173306);浙江省重点研发计划项目(2022C04012) |
| |
| 摘 要： | 复杂化工过程的观测数据往往同时包含非线性和强动态特性，而传统的化工过程软测量方法无法准确提取观测数据的非线性动态特征，以至影响数据建模和质量预报的准确性。提出了一种基于变分自编码器的深度融合特征提取网络(deep fusion features extraction network， DFFEN)。在变分自编码器框架下，通过构建潜隐特征信息传递通道，提取非线性动态潜隐变量。并利用自注意力机制(self-attention)融合关键的隐层信息，优化因信息传递通道过长而导致的潜在特征被遗忘的问题。此外，在后端网络构建潜隐变量和关键质量变量之间的回归模型，以实现关键质量变量的预报。最后，通过数值案例和实际的合成氨过程验证了所提出的DFFEN模型的可行性和有效性。
|
| 关 键 词： | 过程控制 非线性动态建模 神经网络 深度融合特征 合成气 |
| 收稿时间： | 2022-03-07 |
Deep fusion feature extraction network and its application in chemical process soft sensing |
| |
| Le ZHOU,Chengkai SHEN,Chao WU,Beiping HOU,Zhihuan SONG.Deep fusion feature extraction network and its application in chemical process soft sensing[J].Journal of Chemical Industry and Engineering(China),2022,73(7):3156-3165. |
| |
| Authors: | Le ZHOU Chengkai SHEN Chao WU Beiping HOU Zhihuan SONG |
| |
| Affiliation: | 1.School of Automation and Electrical Engineering, Zhejiang University of Science and Technology, Hangzhou 310024, Zhejiang, China;2.State Key Laboratory of Industrial Control Technology, Zhejiang University, Hangzhou 310027, Zhejiang, China |
| |
| Abstract: | The observation data of complex chemical processes often contain both nonlinear and strong dynamic characteristics, and the traditional soft sensing method of chemical process cannot accurately extract the nonlinear dynamic characteristics of the observation data, so as to affect the accuracy of data modeling and quality prediction. In this paper, a deep fusion features extraction network (DFFEN) is proposed. Under the framework of variational auto encoder, the nonlinear dynamic latent variables are extracted by constructing latent feature information transfer channel. In addition, a self-attention mechanism is used to fuse key hidden layer information and optimize the problem that the potential features are forgotten, which is mainly caused by the excessively long information transmission channel. Then, the regression model between latent variables and key quality variables is constructed in the backend network to achieve the prediction of key quality variables. Finally, the feasibility and effectiveness of the proposed DFFEN model are verified by numerical cases and an actual ammonia synthesis process. |
| |
| Keywords: | process control nonlinear dynamic modeling neural networks deep fusion feature extraction syngas |
|
| 点击此处可从《化工学报》浏览原始摘要信息 |
|
点击此处可从《化工学报》下载全文 |
|
