Answers are grounded in this paper and cite their sources.
Soft sensing technology based on semi-supervised learning has gradually become a effective approach for multi-sampling rate industrial scenarios. However, most methods assume similar outputs for analogous inputs, a similarity hypothesis that fails to account for temporal continuity in process industries. To address this issue, we propose a temporal consistency semi-supervised network (TCS-Net) for quality soft sensing of ethylbenzene distillation process with dual sampling rate. TCS-Net initially employs hybrid mutual information for auxiliary feature selection and decomposes the temporal features into trend and fluctuation components. These decomposed temporal sequences are then fed into two parallel neural network prediction channels, with their outputs fused to generate final quality predictions. For unlabeled data under dual sampling rates, a temporal regularization method is designed to exploit latent correlations between labeled and unlabeled data, which is incorporated into the loss function. The effectiveness of the proposed method is validated through a real-world ethylbenzene distillation plant case study.
No takes yet. Share an insight, caveat, or question.
Jin et al. (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: