ADHGR: An optimized adaptive disintegration strategy targeting heterogeneous UAV swarms based on graph reinforcement learning

With the rapid development of unmanned aerial vehicle (UAV) networks, especially in multi-layered and heterogeneous network architectures, it has become increasingly important to ensure that enemy UAV swarms rapidly disintegrate and lose the ability to execute missions under deliberate attacks. Although preliminary studies have been conducted on UAV disintegration strategies, these studies still have non-negligible limitations, such as network homogeneity, irrational self-organization models, and insufficiently intelligent decision-making models. Thus, this work refines the topological model of heterogeneous UAV swarms by considering the communication constraints and dynamic connection strengths of neighbors when modeling the self-organized and self-adaptive manner. Moreover, using penetration theory, this work proposes indicators for evaluating resilience of heterogeneous UAV swarms during disintegration. In addition, an optimized ADHGR approach (Adaptive Disintegration strategy targeting Heterogeneous UAV swarms based on Graph Reinforcement learning) is proposed to efficiently preprocess the heterogeneous graph features and adaptively discover the optimal node removal sequence for network disintegration. Finally, simulation results demonstrate that the proposed model shows superior effectiveness and generalization ability in reducing resilience of heterogeneous UAV swarms compared to conventional targeted disintegration strategies. This model provides a quantitative reference for adaptive disintegration decision-making targeting UAV swarms and supports the development of UAV-swarm related work.