System-of-systems effectiveness evaluation of UAV swarm: A data-driven perspective

System-of-Systems Effectiveness Evaluation (SEE) is crucial for systematically assessing Unmanned Aerial Vehicle Swarms (UAVS). However, due to the strong nonlinear, hierarchical, and emergent characteristics of UAVS, traditional SEE methods based on experience and mathematical analysis are challenging to apply to such complex systems. In this article, we propose a Data-driven SEE framework that leverages deep learning to establish the relationship between system-of-systems indicators and effectiveness, effectively capturing the complex characteristics of UAVS. The system-of-systems indicators are first transformed by two-dimensional loop stacking and then processed by a residual neural network-based evaluation network. Furthermore, to address the practical need to mitigate overestimation in effectiveness evaluation, a skewed loss function is designed to adjust overestimation during the training process. Since training a Data-driven model requires a large number of labeled samples, which are obtained through time-consuming simulation-based annotation, we develop an active learning strategy based on cluster iteration and random greedy sampling. This strategy selects a limited number of high-value samples for simulation labeling while maintaining model accuracy. The experimental results demonstrate that this model can quickly, accurately, and efficiently evaluate the effectiveness of unmanned swarm systems. Additionally, the data-driven SEE framework shows strong potential for UAVS system analysis and optimization.