We propose a novel failure-sentient framework for swarm-based drone delivery services. The framework ensures that those drones that experience a noticeable degradation in their performance (called soft failure) and which are part of a swarm, do not disrupt the successful delivery of packages to a consumer. The framework composes a weighted continual federated learning prediction module to accurately predict the time of failures of individual drones and uptime after failures. These predictions are used to determine the severity of failures at both the drone and swarm levels. We propose a speed-based heuristic algorithm with lookahead optimization to generate an optimal set of services considering failures. Experimental results on real datasets prove the efficiency of our proposed approach in terms of prediction accuracy, delivery times, and execution times.

翻译：我们提出了一种新颖的故障感知框架，用于基于集群的无人机配送服务。该框架确保在集群中性能显著下降（称为软故障）的无人机不会干扰包裹成功送达客户。框架构建了加权持续联邦学习预测模块，以准确预测单架无人机的故障时间及故障后的运行时长。这些预测用于评估无人机层面和集群层面的故障严重程度。我们提出了一种基于速度的启发式算法，结合前向优化策略，在考虑故障的情况下生成最优服务组合。在真实数据集上的实验结果表明，该方法在预测精度、配送时间和执行时间方面具有高效性。