Federated Digital Twins: A Scheduling Approach Based on Temporal Graph Neural Network and Deep Reinforcement Learning
The concept of digital twin (DT), which has supported the optimization of physical system across various industries, now extends into federated digital twin (fDT) to manage complex, interconnected systems. Recent research and standardization efforts in DT have expanded their focus from optimizing in...
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2025-01-01
|
| Series: | IEEE Access |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/10843696/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | The concept of digital twin (DT), which has supported the optimization of physical system across various industries, now extends into federated digital twin (fDT) to manage complex, interconnected systems. Recent research and standardization efforts in DT have expanded their focus from optimizing individual physical twins to coordinating DT functionalities to optimize the services provided by set of physical twins. In fDT-based DT coordination, scheduling techniques that determine when to utilize available DTs efficiently are critical. However, the variable processing speeds of DT functionalities, as traceable from digital twins, present significant challenges in deriving optimal schedules. Traditional scheduling algorithms, which assume static processing speeds for machines (corresponding to DTs), encounter limitations in generating optimal schedules. These limitations are further exacerbated in scenarios that demand the flexibility and scalability inherent to fDT. To address this challenge, this paper proposes the fDT-DT scheduling framework. This framework effectively represents relationships among DTs using temporal heterogeneous graphs and generates flexible and scalable schedules by incorporating temporal graph neural networks and deep reinforcement learning to account for variable processing speeds. Experimental results demonstrate that, in scenarios where multiple DTs exhibit diverse speed profiles, the proposed approach improves makespan by 16.9% and reduces tardiness by 76.2%, demonstrating both flexibility and superior performance. Furthermore, even in untrained scenarios with increasing numbers of fDTs and DTs, it outperforms baseline algorithms on both metrics, confirming its scalability. |
|---|---|
| ISSN: | 2169-3536 |