Knowledge-Enhanced Air-Ground Collaborative Resource Allocation for V2X Networks
Year: 2026
Type: conference-paper
Abstract: Traditional Roadside Unit (RSU) deployments in Vehicle-to-Everything (V2X) networks face high costs and limited coverage in rural or mountainous areas. To address these limitations, this paper proposes an Air-ground cooperative knowledgeenhanced resource allocation framework. Unmanned Aerial Vehicles (UAVs) are introduced as flexible aerial relays to expand RSU service coverage. High-dimensional macro-environmental observations from UAVs are modeled via a Hidden Markov Model (HMM) and compressed into global latent states. To alleviate onboard computational and storage constraints, a lightweight knowledge selection and aggregation mechanism based on Gated Recurrent Units (GRU) is developed, shifting the storage burden back to the ground RSUs. Furthermore, a spatio-temporal prediction-driven cross-region policy pre-warming mechanism utilizing a State Space Model (SSM) is designed to minimize performance fluctuations during handover. Comprehensive simulations across five distinct scenarios demonstrate that the proposed Knowledge-Enhanced Actor-Critic (KE-AC) algorithm achieves convergence within about 80-100 training episodes and outperforms representative DRL baselines in cross-region V2X scenarios, achieving up to 14.3% throughput improvement, 36.7% delay reduction, and 12.7% energy reduction.
Authors: Hao Gong, Minhan Gao, Zhongwei Huang
Institution: Guangdong Polytechnic Normal University
Language: English
Subfield: Electrical and Electronic Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7204209073