具鐵路系統無線鏈路故障預測之最佳化邊緣運算資源分配

Abstract

在5G鐵路通訊系統以及對營運效率和安全性至關重要的各種新興應用情境中，實現視訊偵測的低端到端延遲至關重要。邊緣運算雖具發展潛力，但也帶來了嚴峻挑戰，特別是在受限且動態條件下的高效資源管理問題。此外，鐵路環境固有的高移動性，經常導致不穩定的通訊品質，這是一個關鍵挑戰，可透過無線鏈路故障(RLF)預測模型來主動應對。儘管邊緣運算中的資源管理與RLF預測兩個領域各自存在大量研究，但這兩個關鍵領域在很大程度上仍未整合。再者，既有的資源管理研究常假設為單支線的鐵路系統，從而限制了所提出解決方案在現實多支線環境中的實際適用性和通用性。為了解決這些限制，我們提出了一種新穎的混合式-GGA方法，該方法利用RLF預測結果來動態決定鐵路邊緣運算中的資源分配和任務卸載策略，旨在優化每個時段的整體系統效能。我們的方法透過涵蓋五種不同情境的全面模擬進行評估，並使用真實世界數據集。實證結果突顯了RLF預測的顯著影響，並證明了所提出方法在不同鐵路支線和運作情境下相較於基準方法的卓越效能，從而驗證了其穩健性和通用性。

In the context of 5G-enabled railway communication systems and emerging applications critical for operational efficiency and safety, achieving low end-to-end latency for video detection is essential. While promising, edge computing introduces significant challenges, especially concerning efficient resource management under constrained and dynamic conditions. Furthermore, the inherent high mobility of railway environments frequently results in unstable communication quality, a critical challenge that can be proactively addressed through Radio Link Failure (RLF) prediction models. Despite extensive individual research on resource management in edge computing and on RLF prediction, these two crucial areas largely remain unintegrated. Moreover, prior work in resource management frequently assumes single-branch railway systems, thereby limiting the practical applicability and generality of proposed solutions in realistic multi-branch environments. To address these limitations, we propose a novel Hybrid-GGA approach which leverages RLF prediction results to dynamically determine resource allocation and task offloading strategies for railway edge computing, with the objective of optimizing overall system performance in each timeslot. Our method is evaluated through comprehensive simulations across five distinct scenarios, utilizing a real-world dataset. The empirical results highlight the significant impact of RLF prediction and demonstrate the proposed method's superior effectiveness compared to baselines across diverse railway branches and operational scenarios, thereby validating its robustness and generality.