基於群眾感知系統之賽局激勵與合作任務分配方法;Game-Based Incentive and Cooperative Task Allocation for Crowdsensing System

NCU Institutional Repository > 資訊電機學院 > 通訊工程研究所 > 博碩士論文 > Item 987654321/98039

請使用永久網址來引用或連結此文件: https://ir.lib.ncu.edu.tw/handle/987654321/98039

題名:	基於群眾感知系統之賽局激勵與合作任務分配方法;Game-Based Incentive and Cooperative Task Allocation for Crowdsensing System
作者:	林昆佑;Lin, Kun-Yu
貢獻者:	通訊工程學系
關鍵詞:	群眾感知;激勵機制;賽局理論;合作任務;任務分配;社交網路;Crowdsensing;Incentive mechanism;Game theory;Collaborative multi-user (CMU) task;Task allocation;Social network
日期:	2025-04-21
上傳時間:	2025-10-17 12:16:50 (UTC+8)
出版者:	國立中央大學
摘要:	群眾感知系統（Crowdsensing, CS）是隨著物聯網與用戶為中心的設備之普及，興起的一種新型分散式感知與計算系統。利用廣泛分佈的物聯網設備，包含智慧型手機、可穿戴設備、路側單位和自動駕駛車輛等具有的多感測器、通訊、存儲與計算能力之裝置，群眾感知系統能夠靈活且高效地進行數據收集與分析，應用於交通管理、環境監測、智慧醫療與能源管控等多個應用場景。如今，群眾感知系統在建置和感知服務時，須克服以下幾點問題：如何設計一個強大的用戶招募與激勵機制來擴大用戶基數並提升參與度；如何確保感知資料的數量與品質，特別是在大量且多來源原始資料中過濾低品質的資料；如何整合與管理用戶的資源來處理複雜的感知任務；如何在用戶合作工作環境中，合理分配合作任務並優化用戶資源使用。因此，本論文針對上述群眾感知系統發展面臨的問題，進行基於賽局理論的激勵機制與基於社交關係的合作任務分配之研究。針對群眾感知系統所需的激勵機制，本研究提出了基於Two-stage Stackelberg賽局模型的激勵機制，簡稱為Incentive-G機制，包含資料品質篩選、聲譽評估及投票機制，實現用戶參與行為的有效管理與獎勵分配。首先，Incentive-G機制利用資料相似度來選出不完整或低品質的資料；其次，通過歷史貢獻和聲譽分析來計算用戶的可靠性與信譽評分；最後，基於用戶之間的集體投票來決定資料的有效性，進一步排除惡意用戶。同時，基於Stackelberg賽局模型，該機制能動態調整用戶獎勵並最大化群眾感知系統的總收入。實驗結果顯示，Incentive-G 在資料品質、用戶參與度和獎勵分配效率方面均優於拍賣理論、社交網路與統計分析等經典方法。針對群眾感知系統針對多用戶參與之團隊組合與合作任務分配的需求，本研究提出了一種基於用戶社交網路的群組組成與任務分配機制，稱為 GTA-EDC（Group-based Task Allocation with Effectiveness and Degree Centrality）。該機制設計了一套衡量用戶群組有效性（Effectiveness）與中心性（Degree Centrality）的評估方法，並將其應用於合作任務分配的優化過程。具體而言，GTA-EDC 透過對社交網路中用戶之間的關係進行定量評估，構建出不同類型的用戶群組，包含基本群組、橋接群組、鏈接群組和孤立群組，並基於群組的有效性與中心性指標，選擇最適合執行任務的用戶群組。實驗結果顯示，該方法在降低服務時間、優化執行任務之成本以及提升合作效率方面顯著優於現在的Spanning Tree社群方法、最低成本選擇策略與中心型用戶團隊等合作任務分配方法。總結而言，本論文的研究貢獻在於針對群眾感知系統所需的激勵機制與合作任務分配功能，提出一組新穎與有效的設計方法，透過Incentive-G與GTA-EDC機制的結合，群眾感知系統能夠處理使用者招募、資料品質保障、及多使用者合作任務的分配，進而提升服務的效率與穩定性，同時提供激勵策略與任務分配的聯合優化之創新方法。未來研究將持續拓展本論文的成果，應用於新興的群眾感知服務領域，如多接取邊緣運算（MEC）、行動群眾感知（Mobile Crowdsourcing）及基於社群網路之資訊安全性，這些應用場景中的任務多為具備合作性與高複雜度的群眾感知任務。此外，隱私保護、動態資源分配、及群眾感知系統中使用者間的公平性等關鍵議題，也將是我們未來進一步探討與研究的方向，並提出相應的解決方案。;The pervasion of modern IoT technologies and user-centric devices has paved the way for emerging Crowdsensing (CS) services and applications. A CS system is characterized by its decentralized architecture, which utilizes multi-fold sensing, computing, storage, and networking capabilities of user-centric devices to develop novel applications. Typical sorts of user-centric devices include mobile phones, wearable devices, autonomous vehicles, etc. With the ubiquity of these devices, CS systems can conduct efficient data gathering and analysis in regard to many new applications, e.g., environmental monitoring, energy grid utilization, public safety, social perception, and traffic management, to name a few. Nowadays, several challenges arise from the deployment and utility of CS systems remain unresolved: particularly, how to design robust user recruitment and incentive mechanisms, how to ensure the quality of sensing data, how to deal with complex crowdsensing tasks, how to expand the capabilities of available resource provision in crowd proximity, and how to optimize task allocation in multi-user cooperative contexts. Hence, the study in this dissertation aims to resolve these challenges through two proposed efforts: a game-theory-based incentive mechanism and a social-tie-based cooperative task allocation strategy. For the incentive mechanism adopted by the CS systems, our study proposes a novel solution named Incentive-G, which integrates a two-stage Stackelberg game model with data filtering, reputation evaluation, and user voting mechanisms to effectively manage user participation and reward distribution. The Incentive-G mechanism operates in three key steps: data filtering, reputation-based evaluation, and voting-based decision-making. Initially, incomplete or low-quality data are filtered out. Then, user reliability is assessed based on historical contributions and reputation scores. Finally, collective voting among users determines data validity and enables the exclusion of malicious users. Utilizing the Stackelberg game model, the Incentive-G mechanism dynamically adjusts user rewards to maximize the service provider’s total revenue. Simulation results demonstrate that Incentive-G outperforms existing auction-, social-, and statistics-based methods in terms of data quality, user engagement, and reward allocation efficiency. For the cooperative multi-user task allocation adopted in the CS systems, our study proposes a social network-based group modeling and task allocation mechanism called GTA-EDC (Group-based Task Allocation with Effectiveness and Degree Centrality). This mechanism evaluates user groups’ effectiveness and centrality in a social network to optimize task allocation processes. Specifically, the GTA-EDC mechanism quantifies the relationships between users, forms four types of user groups, i.e., basic, bridging, linked, and isolated groups, and then selects suitable groups for task execution corresponding to their effectiveness and degree centrality. Experimental results indicate that this mechanism significantly reduces service time and energy costs, and meanwhile enhances collaboration efficiency compared to other baseline approaches. In conclusion, the contribution of this dissertation exhibits the feasibility and effectiveness of game-based incentive and cooperative task allocation for the enhancement of CS design and development. Jointly utilizing the Incentive-G and GTA-EDC mechanisms, the CS system is able to deal with user recruitment, data quality assurance, and multi-user cooperative task distribution in CS environments. Future research will continue to develop the efforts of this dissertation in emerging CS application domains, like Multi-Access Edge Computing (MEC), Mobile Crowdsourcing, and Cybersecurity with Social Effects. Wherein, many mission-oriented tasks render in cooperative and complex crowdsensing extents. Additionally, the crucial issues of privacy protection, dynamic resource allocation, and fairness among users in the CS system will be investigated by our further research.
顯示於類別:	[通訊工程研究所] 博碩士論文

文件中的檔案:

檔案	描述	大小	格式	瀏覽次數
index.html		0Kb	HTML	16	檢視/開啟

在NCUIR中所有的資料項目都受到原著作權保護.

社群 sharing

資料載入中.....