無線通訊系統正朝向智慧化、自主化與永續化之網路架構發展，以因應未來第六代（6G）通訊需求。在眾多新興技術中，無人機（UAV）因具備高機動性、彈性部署能力以及提升無線覆蓋與通訊效率之潛力，而受到廣泛關注。同時，能量擷取（Energy Harvesting， EH）與聯邦學習（Federated Learning， FL）亦被視為提升網路永續性與實現分散式人工智慧（AI）服務之關鍵技術。然而，當 UAV 輔助通訊系統與 EH 及 FL 整合後，也衍生出動態資源分配、能源效率最佳化、通訊與學習耦合，以及有限無線資源下之系統可擴展性等挑戰。本論文針對智慧化與永續化 UAV 輔助網路進行研究，並透過通訊、計算與資源管理之聯合設計，探討 EH 與 FL 系統中的關鍵技術。首先，針對 UAV 輔助 EH 通訊系統，本論文建立一套聯合最佳化架構，透過 UAV 軌跡設計、使用者關聯（User Association）以及傳輸功率配置，在能量限制條件下最大化所有使用者之最小總傳輸速率。由於所建立之問題具有高度非凸特性，因此採用逐次凸近似法（Successive Convex Approximation， SCA）與交替最佳化（Alternating Optimization）等方法求解，以獲得高效率之近似最佳解。此外，為提升系統於動態環境中的適應能力，本論文進一步提出結合凸最佳化與強化學習之框架（Convex Optimization-Assisted Reinforcement Learning， CARL），透過離線最佳化與在線學習之整合，使 UAV 系統能夠實現具備良好收斂特性之自適應軌跡與資源管理。接著，本論文將研究延伸至 UAV 輔助聯邦學習系統，並進一步探討通訊與學習流程間之協同設計。為提升系統能源效率，本論文建立一套兼顧通訊與學習效能之聯合設計架構，透過共同最佳化 UAV 軌跡、FL 參與（FL Participation）、無線資源配置以及資料分配，同時滿足聯邦學習之收斂需求。此外，為解決大型 FL 模型所帶來之系統可擴展性問題，本論文進一步提出基於使用者分群（User Grouping）之最佳化架構，以提升通訊可行性與系統擴展能力。所提出之方法能有效在能量消耗、通訊效率與學習效能之間取得平衡。模擬結果顯示，相較於傳統基準方法，本論文所提出之方法於能源效率、通訊可行性以及學習效能方面皆具有顯著改善。整體而言，本論文建立一套涵蓋 EH 通訊、智慧資源管理與聯邦學習之統一研究框架，為未來 AI 驅動之 6G UAV 輔助無線通訊系統提供重要之研究洞見與設計方向。;With the evolution of wireless communication systems toward intelligent， autonomous， and sustainable network architectures， emerging technologies are being actively explored to meet the requirements of future sixth-generation (6G) communications. Among these technologies， unmanned aerial vehicles (UAVs) have attracted significant attention due to their high mobility， flexible deployment capability， and potential to enhance wireless coverage and communication efficiency. Meanwhile， energy harvesting (EH) and federated learning (FL) have also been recognized as key enablers for improving network sustainability and realizing distributed artificial intelligence (AI) services. However， integrating UAV-assisted communications with EH and FL systems introduces several new challenges， including dynamic resource allocation， energy-efficient optimization， communication-learning coupling， and system scalability under limited wireless resources.
This dissertation investigates intelligent and sustainable UAV-assisted networks through the joint design of communication， computation， and resource management mechanisms， with a particular focus on EH and FL systems. First， for UAV-assisted EH communication systems， a joint optimization framework is developed by jointly considering UAV trajectory design， user association， and transmit power allocation to maximize the minimum achievable throughput among all users under energy constraints. Since the formulated problem is highly non-convex， successive convex approximation (SCA) and alternating optimization techniques are employed to obtain computationally efficient near-optimal solutions. Furthermore， to enhance system adaptability in dynamic environments， a Convex Optimization-Assisted Reinforcement Learning (CARL) framework is proposed by integrating offline optimization with online learning， enabling adaptive UAV trajectory and resource management with favorable convergence performance.
The dissertation further extends the study to UAV-assisted federated learning systems by jointly considering communication and learning processes. To improve system energy efficiency， a joint design framework is proposed to jointly optimize UAV trajectory， FL participation， wireless resource allocation， and data allocation while satisfying the convergence requirements of federated learning. In addition， to address the scalability challenges caused by large-scale FL models， a user grouping-based optimization framework is further developed to enhance communication feasibility and system scalability. The proposed approaches effectively balance energy consumption， communication efficiency， and learning performance.
Simulation results demonstrate that the proposed methods significantly outperform conventional benchmark schemes in terms of energy efficiency， communication feasibility， and learning performance. Overall， this dissertation establishes a unified research framework integrating EH communications， intelligent resource management， and federated learning， providing important insights and design guidelines for future AI-driven 6G UAV-assisted wireless communication systems.
