AGI-Inspired Digital Twin Framework for UAV-BS Deployment in Disaster Communication Recovery

Unmanned aerial vehicle base stations (UAV-BSs) have shown significant promise in enabling rapid communication recovery in disaster scenarios. However, existing approaches are often task-specific or heuristic-based, lacking the adaptability to handle UAV-BS deployment in uncertain, dynamic disaster scenarios. To address these issues, we propose an artificial general intelligence (AGI) inspired digital twin (DT) framework for UAV-BS deployment, aligning the design of DT with AGI’s closed-loop architecture consisting of perception, modeling, and action planning. Specifically, the proposed DT framework includes a predictor and an optimizer. The predictor, implemented by a graph neural network (GNN) and multilayer perceptron (MLP), performs perception and modeling of real UAV-BS network states and predicts service satisfaction. The optimizer, developed based on hierarchical reinforcement learning (HRL), performs action planning by adaptively determining the number and trajectories of UAV-BSs, subject to multiple network performance and operational constraints. A self-evolving loop is formed through online learning, which continuously updates the DT using feedback from the real UAV-BS network. Extensive simulation experiments are conducted to validate the effectiveness of the proposed DT, with the predictor achieving 98.49% for average accuracy on service satisfaction and the optimizer reducing training time by 63.73% compared to multi-level deep Q-learning. Additionally, optimized UAV-BS deployment decisions can be generated within 0.086 seconds when the disaster environment condition changes.