{"title":"Attention-augmented multi-agent collaboration for Smart Industrial Internet of Things task offloading","authors":"Yihang Wang , Shengchao Su , Yiwang Wang","doi":"10.1016/j.iot.2025.101572","DOIUrl":null,"url":null,"abstract":"<div><div>The integration of Multi-access Edge Computing (MEC) technology within the Smart Industrial Internet of Things (SIIoT) ecosystem can significantly enhance both computational and storage capabilities. This advancement facilitates improved data processing and a more efficient utilization of resources in industrial applications. However, the high density of devices typical of SIIoT environments often presents several challenges, including a low success rate for task offloading, increased latency, higher energy consumption, and the risk of overloading edge servers. This paper addresses these challenges by treating Smart Devices (SDs) as agents and proposing a collaborative multi-agent task offloading strategy. A computational offloading model has been developed to minimize delayed energy consumption, which is then formulated as a Multi-Agent Partially Observable Markov Decision Process (MAPOMDP) featuring a hybrid action space composed of discrete and continuous elements. An attention mechanism is introduced to tackle the complex competition for edge server resources among SDs during the offloading process, enabling the observation of the actions and states of other devices within the system. A Prioritized Experience Replay (PER) mechanism is employed to optimize the training process. A Multi-Agent Attention Deep Reinforcement Learning (MA2DRL) algorithm is proposed to improve computational task offloading. Experimental results demonstrate that the proposed algorithm outperforms other comparative algorithms regarding task offloading latency, average energy consumption, offloading success rate, and server load variance.</div></div>","PeriodicalId":29968,"journal":{"name":"Internet of Things","volume":"31 ","pages":"Article 101572"},"PeriodicalIF":6.0000,"publicationDate":"2025-03-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Internet of Things","FirstCategoryId":"94","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S254266052500085X","RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"COMPUTER SCIENCE, INFORMATION SYSTEMS","Score":null,"Total":0}
引用次数: 0
Abstract
The integration of Multi-access Edge Computing (MEC) technology within the Smart Industrial Internet of Things (SIIoT) ecosystem can significantly enhance both computational and storage capabilities. This advancement facilitates improved data processing and a more efficient utilization of resources in industrial applications. However, the high density of devices typical of SIIoT environments often presents several challenges, including a low success rate for task offloading, increased latency, higher energy consumption, and the risk of overloading edge servers. This paper addresses these challenges by treating Smart Devices (SDs) as agents and proposing a collaborative multi-agent task offloading strategy. A computational offloading model has been developed to minimize delayed energy consumption, which is then formulated as a Multi-Agent Partially Observable Markov Decision Process (MAPOMDP) featuring a hybrid action space composed of discrete and continuous elements. An attention mechanism is introduced to tackle the complex competition for edge server resources among SDs during the offloading process, enabling the observation of the actions and states of other devices within the system. A Prioritized Experience Replay (PER) mechanism is employed to optimize the training process. A Multi-Agent Attention Deep Reinforcement Learning (MA2DRL) algorithm is proposed to improve computational task offloading. Experimental results demonstrate that the proposed algorithm outperforms other comparative algorithms regarding task offloading latency, average energy consumption, offloading success rate, and server load variance.
期刊介绍:
Internet of Things; Engineering Cyber Physical Human Systems is a comprehensive journal encouraging cross collaboration between researchers, engineers and practitioners in the field of IoT & Cyber Physical Human Systems. The journal offers a unique platform to exchange scientific information on the entire breadth of technology, science, and societal applications of the IoT.
The journal will place a high priority on timely publication, and provide a home for high quality.
Furthermore, IOT is interested in publishing topical Special Issues on any aspect of IOT.