基于条件生成对抗网络的扇区复杂度评估

扇区复杂度作为管制员工作负荷和动态空域配置的重要参考依据，需要事先准确地对其进行评估。本文针对有监督复杂度数据集存在的小样本问题，提出基于条件生成对抗网络的扇区复杂度评估框架。首先，构建交通流量、航空器性能和潜在冲突这3类复杂度指标，并结合主观复杂度等级得到标定样本；其次，利用条件生成对抗网络设计有标记样本生成算法，获得增广数据集；最后，分别采用逻辑回归、支持向量机和随机森林算法建立复杂度评估模型。以中南区域扇区为例，从定性和定量的视角验证生成样本的有效性，并在多种训练集配置下对比各模型评估结果。研究结果表明：条件生成对抗网络在200次迭代后逐步收敛至稳定；生成样本与真实样本的绝大多数指标在均值上的相对误差小于5%，在标准差上的相对误差大于5%；在多分类评价指标下，增广数据集对3种模型整体评估精度分别提升11.77%、11.04%和8.34%。本文提出的评估框架可以在有限数据条件下提高样本多样性，是解决扇区复杂度评估问题的一种有效方法。

Improved Model and Algorithm for Optimizing Collaborative Trajectory Options Program

To optimize the collaborative trajectory options program (CTOP), the Gini coefficient is firstly introduced to define a new metric of equity. A bi- objective integer nonlinear programming model is formulated considering the performance of efficiency and equity. An improved genetic algorithm is applied to solve the model, which uses an array of integers representing a flight priority order to code the chromosome and adds a process to choose the satisfactory solution. According to the simulation, the final satisfactory solution based on the improved algorithm brings a 9.3% increase in the airspace operating efficiency and a 33.7% increase in the airlines' equity, compared with the solution solved by the current practical algorithm. The result shows the improved algorithm produces the true Pareto frontiers quickly, and the final satisfactory solution both improves the efficiency and equity significantly.