.png){kind=icon}
.png){kind=icon}
Return
CRTSE: Clustering and Reinforcement Learning Based Task Scheduling Algorithm for Edge Computing
.png){kind=icon}
PRE
AI
DOI：10.1109/TCC.2025.3640958.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Edge computing can overcome many shortcomings of traditional cloud computing and provide high quality computing services. However, it need to face the challenges of node heterogeneity and task diversity, which leads to the performance of edge computing systems relying on proper task scheduling. Existing task scheduling algorithms are usually designed based on mathematical models that are closely related to the internal details of edge computing systems, resulting in poor universality and limited quality of scheduling decisions. In this paper, we address these issues by proposing a Clustering and Reinforcement Learning Based Task Scheduling Algorithm for Edge Computing (CRTSE), which aims to shorten the task processing time and improve the energy efficiency of the system. The algorithm uses Adaptive Graph Auto-Encoder (AdaGAE) based clustering algorithm to cluster computing tasks and classify the computing tasks submitted to the edge computing system based on the clustering results. When making scheduling decisions, an independent deep reinforcement learning model is used for each class of computing tasks to obtain targeted scheduling preference information, and the scheduling decision is made based on these preference information. CRTSE possesses good universality due to the feature of not relying on mathematical models that are closely related to the internal details of edge computing systems, and has the ability to adapt well to diverse computing tasks and to learn continuously during the interaction with the system. Simulation experiments based on real data show that CRTSE can shorten the average task processing time of the edge computing system by up to 56.00% and reduce the average energy consumption of the edge computing system by up to 16.10% compared to existing excellent scheduling algorithms.
Keywords:
Edge computing
cloud computing
task scheduling
clustering
machine learning
deep reinforcement learning
Journal
I
IF:
5
Papers:
1.8K
Citations:
4.3K