Online Prediction of Industrial Process Index: An Edge-Cloud Collaborated Approach

In this article, an edge-cloud collaborative prediction algorithm is proposed for the problem of industrial process index forecasting. In the proposed algorithm, the cloud-based models are updated periodically, while the edge-side models are updated multiple times using the latest data during each c...

Full description

Saved in:
Bibliographic Details
Published in:IEEE transactions on industrial electronics (1982) Vol. 71; no. 9; pp. 11419 - 11428
Main Authors: Lu, Shaowen, Zhang, Yajun, Wu, Jinfeng
Format: Journal Article
Language:English
Published: New York IEEE 01.09.2024
The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
Subjects:
Algorithms
Cloud computing
Collaboration
Computation offloading
Computational modeling
Data models
edge-cloud collaboration
industrial Internet of Things
industrial process index forecasting
operation optimization control
Potential oscillations
Predictive models
Production
Synchronism
Synchronization
ISSN:0278-0046, 1557-9948
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:In this article, an edge-cloud collaborative prediction algorithm is proposed for the problem of industrial process index forecasting. In the proposed algorithm, the cloud-based models are updated periodically, while the edge-side models are updated multiple times using the latest data during each cloud update cycle. We use a linear model for the edge-side and a nonlinear long short term memory model for the cloud-based model and design a novel synchronization mechanism to update the models of both sides alternatively. In order to mitigate potential oscillations caused by frequent updates to the edge-side models, two additional coefficients are introduced. We provide a formal analysis of the convergence of prediction errors. The proposed algorithm is validated using both simulation and real industrial data. The results show that the forecast accuracy and stability are remarkably improved compared to other methods.
Bibliography:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:0278-0046
1557-9948
DOI:10.1109/TIE.2023.3335470