Novel network framework proposed for operation state prediction and management

Researchers have proposed a multivariate time-series prediction framework based on a deep hybrid network and innovatively introduced noise elimination technology in deep learning modeling.

Their findings were published in IEEE Transactions on Industrial Informatics.

Industrial process systems operate all over the place. But do you know what is happening inside them? Scientists and engineers know about the hidden complexities of physical-chemical reactions, heat-mass transfer, and thermal-hydraulic processes that are taking place every second. For years, researchers have been trying in different ways to monitor, diagnose and predict normal and safe operations.

However, accurate prediction is no walk in the park in any situation, let alone those giant and complicated systems. With the rapid development of artificial intelligence technology, soft measurement rises to cope, which collects data from neighboring somewhere. Considering that some data is not directly available, to figure out a result after inputting those indirect data into modeling. Considering that some of the data is not directly available, then indirect data can be input into the modeling, and the result is calculated.

"So, when it comes to modeling and calculation, data is vital," said YAO Yuantao, a research fellow of GE Daochuan's group from the Hefei Institutes of Physical Science (HFIPS) of the Chinese Academy of Sciences (CAS), also the first author of the study.

"When our goal is accuracy, first we have to figure out what creates uncertainty. We found the specific noise," YAO talked about what their study started with.

The team found that the specific noise generated by different component monitors that are applied in system operation to collect data would cause uncertainty in the data and then prediction accuracy is reduced as a result.

"We focused our work on how to remove data uncertainty caused by noise inside," said YAO.

They proposed a deep hybrid network-based multivariate time-series prediction framework, which consists of three modules working together to obtain certain key monitoring industrial process safety metrics.

"We built module for parameter variable selection, then feature selection function for input monitoring process data can be realized," YAO described the modules they integrated into their framework.

Besides, they also set up a residual elimination module with a convolution structure to eliminate the data noise uncertainty and a prediction module with a time-distributed gated recurrent unit for trend prediction of the monitored index.

Together, these three modules provide a combination of data noise uncertainty elimination and time-series prediction. They can effectively reduce the uncertainty caused by different degrees and types of noise.

In addition, using the redundant information in the same type of monitoring data between adjacent nodes is helpful to improve the prediction performance.

"It is largely due to the residual elimination design in the framework," said YAO, "We tested it on the proposed nuclear platform dataset benchmark with good results."

This work, according to the team, provides a reference for other industrial process systems.

More information:
Yuantao Yao et al, Multivariate Time Series Prediction in Industrial Processes via a Deep Hybrid Network under Data Uncertainty, IEEE Transactions on Industrial Informatics (2022). DOI: 10.1109/TII.2022.3198670

Provided by Chinese Academy of Sciences