Unmanned aerial systems (UASs) are continuing to proliferate. Quantitative risk assessment for UAS operations, both a priori and during the operation, are necessary for governing authorities and insurance companies to understand the risks and properly approve operations and assign insurance premiums, respectively. In this paper, the problem of UAS risk analysis and decision making is treated through a novel application of Dempster–Shafer (DS) networks using auto-updating transition matrices. This method was motivated by the results of the 2018 UAS Safety Symposium held at the Georgia Institute of Technology, which was conducted as part of the research detailed in this paper. The paper describes training a DS network based on simulated operation data, testing the capabilities of the trained network to make real-time decisions on a small UAS against a baseline system in a representative mission, and exploring how this system would extend to a more inclusive UAS ecosystem. Conclusions are drawn with respect to the research performed, and additional research directions are proposed.

无人机系统（UAS）仍在继续扩散。在先验过程中和在操作过程中，对UAS操作进行定量风险评估对于管理机构和保险公司分别了解风险并适当批准操作并分配保险费是必不可少的。在本文中，通过使用自动更新过渡矩阵的Dempster-Shafer（DS）网络的新颖应用来解决UAS风险分析和决策问题。该方法是由佐治亚理工学院举行的2018年UAS安全研讨会的结果所激发的，该研讨会是本文详细研究的一部分。本文介绍了基于模拟操作数据训练DS网络的方法，测试具有代表性的任务中经过训练的网络的能力，以针对基准系统上的小型UAS做出实时决策，并探索该系统将如何扩展到更具包容性的UAS生态系统。针对所进行的研究得出结论，并提出了其他研究方向。