文献类型：会议论文

中文题名：SFENet: Deep Neural Network with Separated Feature Extraction for Erythromycin Fermentation Production Prediction

作者：Zhaoran Liu[1];Chao Li[2];Xiwei Tian[2];Yizhi Cao[1];Zijian Tian[1];Xinggao Liu[1]

机构：[1]College of Control Science and Engineering, Zhejiang University;[2]School of Biotechnology, East China University of Science and Technology

会议论文集：第35届中国过程控制会议论文集

会议日期：20240725

会议地点：中国海南三亚

语种：英文

中文关键词：Erythromycin fermentation production;Prediction models;Neural networks

摘要：Large-scale erythromycin fermentation production is a complex process, and it is challenging to optimize the fed-batch strategy by only relying on manual control. Therefore, a prediction model is proposed to automatically predict key indicators in the erythromycin fermentation production process to assist human experts in formulating fed-batch strategy. The proposed model, SFENet, fully considers the time lag effect in monitoring data and the feature confusion problem in representation learning. The architecture of SFENet enables it to simultaneously learn temporal and variable correlations from the data and adaptively optimize the graph structure during the training process. Experimental results on real-world datasets show that SFENet reduces the prediction error by an average of 10% compared to previous state-of-the-art models and achieves excellent performance in both short-term and long-term scenarios. The proposed model is capable of being deployed in fermenters to assist in improving erythromycin production.