CIESC Journal ›› 2023, Vol. 74 ›› Issue (6): 2503-2521.DOI: 10.11949/0438-1157.20230332

• Process system engineering • Previous Articles     Next Articles

Fault monitoring of fermentation process based on attention dynamic convolutional autoencoder

Xuejin GAO1,2,3,4(), Yuzhuo YAO1,2,3,4, Huayun HAN1,2,3,4(), Yongsheng QI5   

  1. 1.Faculty of Information Technology, Beijing University of Technology, Beijing 100124, China
    2.Engineering Research Center of Digital Community, Ministry of Education, Beijing 100124, China
    3.Beijing Laboratory for Urban Mass Transit, Beijing 100124, China
    4.Beijing Key Laboratory of Computational Intelligence and Intelligent System, Beijing 100124, China
    5.School of Electric Power, Inner Mongolia University of Technology, Hohhot 010051, Inner Mongolia, China
  • Received:2023-04-03 Revised:2023-05-06 Online:2023-07-27 Published:2023-06-05
  • Contact: Huayun HAN

基于注意力动态卷积自编码器的发酵过程故障监测

高学金1,2,3,4(), 姚玉卓1,2,3,4, 韩华云1,2,3,4(), 齐咏生5   

  1. 1.北京工业大学信息学部,北京 100124
    2.数字社区教育部工程研究中心,北京 100124
    3.城市轨道交通北京实验室,北京 100124
    4.计算智能与智能系统北京市重点实验室,北京 100124
    5.内蒙古工业大学电力学院,内蒙古 呼和浩特 010051
  • 通讯作者: 韩华云
  • 作者简介:高学金(1973—),男,博士,教授,gaoxuejin@bjut.edu.cn
  • 基金资助:
    北京市自然科学基金项目(4192011)

Abstract:

The status monitoring of the fermentation process plays a vital role in timely detection of various abnormal faults. However, due to the nonlinear characteristics of the fermentation process data, it is difficult to extract feature information, which increases the difficulty of fault monitoring. In order to solve the above problems, an attention dynamic convolutional autoencoder (ADCAE) based fault monitoring method for fermentation process is proposed. Firstly, a dynamic convolution structure is designed, which can extract low-level features using large-size convolution kernels in the shallow layer, and extract high-level features using small-size convolution kernels in the deep layer, thereby broadening the scope of model feature learning scale. Secondly, a channel convolutional attention (CCA) module is designed, which can extract the nonlinear features of input from different scales, and can be better extract local features in the process of converting channel vectors into weights, which improves the ability to pay attention to effective information. Finally, the dynamic convolution structure and CCA module are integrated into the convolutional autoencoder, so that the model can effectively capture the nonlinear relationship in the variables, so as to better cope with the problem of fault monitoring in the fermentation process. The feasibility of the method was verified by using the simulation platform of penicillin fermentation process and the actual production data of Escherichia coli, and the results showed that the method had good fault monitoring performance.

Key words: fermentation, algorithm, nonlinear, fault monitoring, neural networks, attention mechanism, convolutional autoencoder

摘要:

发酵过程的状态监测对于及时发现各类异常故障起到了至关重要的作用。然而,由于发酵过程数据呈现非线性特性,导致在提取特征信息时存在困难,增加了故障监测的难度。为了解决上述问题,提出了一种基于注意力动态卷积自编码器(attention dynamic convolutional autoencoder, ADCAE)的发酵过程故障监测方法。首先,设计了一种动态卷积结构(dynamic convolution structure),该结构可以在浅层使用大尺寸卷积核提取低级特征,在深层使用小尺寸卷积核提取高级特征,从而拓宽了模型特征学习的尺度;其次,设计了一种通道卷积注意力(channel convolutional attention, CCA)模块,该模块能够从不同尺度提取输入的非线性特征,并且在通道向量转化为权重的过程中可以更好地提取局部特征,提高了对有效信息的关注能力;最后,将动态卷积结构与CCA模块融入卷积自编码器中,使模型能够有效地捕获变量中的非线性关系,从而更好地应对发酵过程中的故障监测问题。利用青霉素发酵过程仿真平台和大肠埃希菌实际生产数据对该方法的可行性进行了验证,结果表明该方法具有较好的故障监测性能。

关键词: 发酵, 算法, 非线性, 故障监测, 神经网络, 注意力机制, 卷积自编码器

CLC Number: