A method for generating high-risk scenarios of UAV swarms based on the degree of causal contribution

To ensure the stable performance of UAV swarms in complex mission environments, the system must undergo rigorous test and evaluation procedures under large-scale scenarios that encompass diverse environmental and operational constraints. High-risk scenarios, characterized by potential or imminent collisions among UAV swarm members, represent critical testing conditions. However, such scenarios typically exhibit a long-tail distribution and occur infrequently, which complicates their identification and capture. Furthermore, high-risk events result from the interplay of multiple contributing factors, making it challenging to quantify the magnitude of individual influences in a manner that facilitates accelerated scenario generation. To address these challenges, this study introduces a high-risk scenario generation framework that integrates surrogate modelling with causal contribution analysis. Drawing upon counterfactual causal theory, a Degree of Causal Contribution (DCC) metric is developed to quantify the extent of disturbance introduced by individual factors contributing to high-risk scenarios. A Gaussian Process Regression (GPR) surrogate model is then constructed to predict partial DCC values across various experimental conditions, thereby guiding the targeted generation of high-risk scenarios. A DCC-based uniqueness criterion is further implemented to minimize redundancy in generated scenarios. Experimental results indicate that the GPR model exhibits strong predictive accuracy and fitting performance. Compared to approaches that rely solely on DCC metrics, the proposed framework demonstrates enhanced capability in generating a broader set of unique high-risk scenarios, thereby offering a more robust methodology for UAV swarm testing.