Attention-Driven Multi-Objective Optimization for AAV-Assisted Cellular Networks

With the explosive growth of users and data-hungry applications, cellular networks are increasingly turning to Autonomous Aerial Vehicles (AAVs) as agile, on-demand aerial base stations. AAV-assisted cellular networks face challenges in multi-objective optimization under dynamic environments, as conventional fixed-weighting schemes fail to adapt to real-time conditions. To address this limitation, we propose a Multi-objective Attention-based Proximal Policy Optimization (MOAPPO) framework, which leverages a multi-head attention mechanism to dynamically generate context-aware objective weights. These adaptive weights guide an enhanced k-means user clustering process and a Proximal Policy Optimization (PPO)-based strategy for joint AAV trajectory planning and power allocation under parameterized Non-Orthogonal Multiple Access (NOMA) control. By integrating all components into a unified closed-loop architecture, the framework continuously adapts to the dynamic mobility of both AAVs and users. Extensive experimental results demonstrate that MOAPPO outperforms existing baselines in terms of system performance, while ensuring improved load balancing and stable convergence behavior.