Cooperative Edge Content Caching With Popularity Prediction in UAV-Assisted Vehicular Networks

Edge content caching is a key enabler of low-latency multimedia delivery (e.g., short videos and images) in unmanned aerial vehicle (UAV)-assisted vehicular networks (UVNs). Nevertheless, UVNs still face substantial challenges in achieving efficient delivery under stringent latency constraints, mainly due to UAVs' limited onboard storage and the highly dynamic nature of content popularity. To address these challenges, this paper proposes a cooperative edge content caching scheme for UVNs. Specifically, we first present a hierarchical caching framework driven by popularity prediction. Highly popular content is replicated across all UAVs, whereas low-to-medium-popularity content is cooperatively distributed among them. To forecast popularity, we design a two-module predictor that couples a long short-term memory (LSTM) network, which captures temporal request patterns, with an attention mechanism that adaptively reweights salient contextual features. We further model cooperative interactions among UAVs as a coalition-formation game that incentivizes collaborative caching of low-to-medium-popularity content. Within this game, we develop a stable coalition-partition formation algorithm to enable mutually beneficial cooperation, and a dynamic-programming-based greedy algorithm to derive near-optimal caching decisions for UAVs within each coalition. Finally, extensive experiments demonstrate that the proposed approach substantially improves popularity-prediction accuracy and effectively incentivizes cooperative caching, thereby enhancing overall content-delivery efficiency in UVNs.