Task-driven Decentralized Decision-making Method for UAV Swarm with Behavior Regulation under Communication Delay

Classic swarm models attribute behavioral emergence to individual-level local interactions, enabling unmanned aerial vehicle (UAV) swarms to execute basic tasks. However, communication delay and packet loss caused by electromagnetic interference or link fluctuations trigger decision-making biases, severely limiting swarm adaptability to complex multi-task scenarios. To address this challenge, this study proposes a Delay Prediction-based Task-driven Swarm Decision-making Method (DP-TDDM), comprising three core modules: a front-end trajectory prediction module, a neighbors reselection module, and a back-end task-driven individual-level decision-making module. In the trajectory prediction module, a residual-corrected non-equidistant grey prediction model is established, which achieves high-precision small-sample prediction by dynamically updating the delay queue via a sliding window mechanism. In the neighbors reselection module, a quantitative criterion integrating information timeliness and flight safety is used to perform reselection and assign differentiated state weights to the new neighbors, thereby optimizing decision-making information source quality systematically. In the individual-level decision-making module, an orderly rule-behavior mapping is built to realize event-triggered, task-driven dynamic decision-making, which ensures conflict-free behavioral switching while preserving collective motion robustness. Based on a high-fidelity hardware-in-the-loop (HIL) simulation system, comprehensive performance comparisons are conducted between DP-TDDM and classic Intelligent Self-Organized Algorithm (ISOA), Weak Information Interaction UAV Swarm Model (WIIUSM) and their variants in multi-task scenarios. Results verify DP-TDDM’s effectiveness, robustness, and scalability, while additional experiments on delay prediction accuracy and behavior regulation effectiveness further corroborate its superiority. This study provides new insights based on classical swarm models and enhances the task adaptability of UAV swarms under communication delay.