Soft sensor modeling method for Pichia pastoris fermentation process based on domain adaptation ensemble LSTM

Soft sensor modeling method for Pichia pastoris fermentation process based on domain adaptation ensemble LSTM

To address the challenges of limited labeled sample information and the inability to effectively utilize public information across different fermentation batches in Pichia pastoris fermentation process, this paper propose a soft-sensing modeling method based on domain adaptation ensemble LSTM. First...

Full description

Saved in:
Bibliographic Details
Main Authors: Bo Wang, Jun Wei, Yongxian Song, Hui Jiang
Format: Article
Language:English
Published: Elsevier 2025-10-01
Series:Alexandria Engineering Journal
Subjects:
Soft sensor model
Domain adaptation
Ensemble learning
LSTM
Online Access:http://www.sciencedirect.com/science/article/pii/S1110016825007070
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:To address the challenges of limited labeled sample information and the inability to effectively utilize public information across different fermentation batches in Pichia pastoris fermentation process, this paper propose a soft-sensing modeling method based on domain adaptation ensemble LSTM. Firstly, in order to effectively utilize unlabeled data and improve the generalization ability of the model, autoencoder technology is adopted to extract dynamic feature information of the fermentation process from unlabeled data using LSTM encoder decoder. Secondly, based on the LSTM encoder decoder framework, LSTM deep neural networks are used to establish corresponding sub-models for each fermentation batch. Finally, during the training of each LSTM sub-model, a domain adaptive ensemble learning strategy is introduced to incorporate the effective information of other fermentation batches under different operating conditions into the loss function of the sub model, in order to reduce the impact of distribution differences between different fermentation batches on the sub-model. The simulation results show that the mentioned method has the preponderance of timely prediction and high prediction accuracy, which validates the effectiveness and practicality of the method.
ISSN:1110-0168

Similar Items