移動群眾感知平台上適用於反向拍賣模型的DQN參與者選擇方法;DQN-based Participant Selection for Reverse Auction in Mobile Crowdsensing Platform

NCU Institutional Repository > 資訊電機學院 > 通訊工程研究所 > 博碩士論文 > Item 987654321/95396

jsp.display-item.identifier=請使用永久網址來引用或連結此文件: http://ir.lib.ncu.edu.tw/handle/987654321/95396

题名:	移動群眾感知平台上適用於反向拍賣模型的DQN參與者選擇方法;DQN-based Participant Selection for Reverse Auction in Mobile Crowdsensing Platform
作者:	孫全賢;SUN, CHUAN-HSIEN
贡献者:	通訊工程學系
关键词:	移動群眾感知;強化學習;參與者選擇;拍賣理論;用戶特徵;行動計算;物聯網;Mobile Crowdsensing;Reinforcement Learning;Participant selection;Auction theory;User characteristic;Mobile computing;Internet of Things(IoT)
日期:	2024-08-20
上传时间:	2024-10-09 16:45:56 (UTC+8)
出版者:	國立中央大學
摘要:	移動群眾感知技術通過集體努力，利用來自大量移動用戶的數據，顯著提高了環境中數據利用率。在這個過程中，移動用戶參與感知任務並獲取相應的獎勵。然而，參與者的感測設備和感測行為通常不專業進而導致感測品質低或甚至是提交無用的數據。在此背景下，本文提出了一種移動群眾感知平台上適用於反向拍賣模型的參與者選擇機制，旨在穩定地招募移動參與者，提高感知數據的可靠性和質量。這一機制透過反向拍賣使平台能夠招募到最合適的參與者。平台會分析傳感數據，確定移動參與者的聲稱成本以及實際成本，以確保數據的質量和可靠性。該機制透過強化學習（DRL）根據任務需求和環境變化動態選擇最合適的參與者，從而避免收集過多冗餘數據。強化學習不斷學習，將任務分配給最適合的參與者，從而在有限預算下提高平台的利潤並維持穩定性。效能結果表明，我們的機制能夠使移動群眾感知平台在長期任務分配過程中最大化整體任務完成度的同時最大化平台的收益，也能有效防範異常參與者所帶來的影響。;Mobile crowdsensing (MCS) technology leverages collective efforts and data from a large number of mobile participants to significantly enhance data utilization efficiency in various environments. In this process, mobile participants complete tasks and receive corresponding rewards. However, the sensing devices and behaviors of participants are often unprofessional, leading to poor sensing quality or even submission of useless data.This article proposes a participant selection mechanism suitable for reverse auction models on MCS platforms. The aim is to stably recruit mobile participants, improve the reliability and quality of sensed data, and prevent the influence of abnormal participants. Through reverse auctions, the platform can recruit the most suitable participants. The platform analyzes the sensed data to determine the claimed and actual costs of mobile participants, ensuring data quality and reliability. The mechanism uses deep reinforcement learning (DRL) to dynamically select the most appropriate participants based on task requirements and environmental changes, thus avoiding the collection of excessive redundant data. Reinforcement learning continuously learns and allocates tasks to the most suitable participants, thereby increasing the platform′s profit and maintaining stability under limited budget. Performance results manifest that our mechanism enables the MCS platform to maximize overall task completion rate and platform revenue during long-term task allocation processes, effectively mitigating the impact of abnormal participants.
显示于类别:	[通訊工程研究所] 博碩士論文

文件中的档案:

档案	描述	大小	格式	浏览次数
index.html		0Kb	HTML	26	检视/开启

在NCUIR中所有的数据项都受到原著作权保护.

社群 sharing

数据加载中.....