.png){kind=icon}
.png){kind=icon}
Return
Coupled Temporal and Relevant Feature Prediction for Process Quality: A Graph-Informed Gated Recurrent Unit
.png){kind=icon}
PRE
AI
W
Y
R
Y
W
J
DOI：10.1109/tii.2026.3678504.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Process industries typically exhibit highly coupled process variables and fluctuations in raw materials, making accurate quality prediction challenging. Conventional methods rely on either parameter-quality relevant modeling or temporal modeling. However, these methods often overlook the dynamic coupling effects, thereby limiting their predictive capabilities. To address this, we present an integrated framework that combines graph-based computation with gated recurrent units (GRUs) to explicitly model dynamic coupling in quality prediction. In the proposed framework, process parameters and quality indicators are represented as graph nodes, while interactions among variables are encoded as edges. Production data are used to initialize the node features, and graph neural networks are used to extract relevant representations. These representations are then passed to a graph-informed GRU (GIGRU) cell, an enhanced GRU architecture that incorporates internal mechanisms for feature selection and state updating. By integrating coupling information with memory and coupling update gates, the architecture enables joint modeling of relevant and temporal dependencies, providing stronger representational power than traditional methods. The proposed approach was evaluated on data from a tobacco production line. Experimental results showed that our GIGRU model achieved a mean absolute error of 0.061, a mean squared error of 0.605, and a coefficient of determination of 0.942, outperforming the other prediction methods compared. Ablation studies further confirm the model’s improved ability to capture both temporal dynamics and relational coupling in process data. Overall, the proposed approach enhances prediction accuracy and robustness, offering more reliable support for quality monitoring and optimization in process industries.
Keywords:
Coupling prediction
graph neural networks
process manufacturing
quality prediction
time series prediction
Journal
IF:
9.9
Papers:
8.3K
Citations:
6.0W
Organization
Cited Papers