Accurate prediction of real-time airport capacity, a.k.a. airport acceptance rates (AARs), is key to enabling efficient air traffic flow management. AARs are dependent on selected runway configurations and both are affected by weather conditions. Although there have been studies tackling on the prediction of AARs or runway configurations or both, the prediction accuracy is relatively low and only single airport is considered. This study presents a data-driven deep-learning framework for predicting both runway configurations and AARs to support efficient air traffic management for complex multi-airport systems. The two major contributions from this work are 1) the proposed model uses assembled gridded weather forecast for the terminal airspace instead of an isolated station-based terminal weather forecast, and 2) the model captures the operational interdependency aspects inherent in the parameter learning process so that proposed modeling framework can predict both runway configuration and AARs simultaneously with higher accuracy. The proposed method is demonstrated with a numerical experiment taking three major airports in New York Metroplex as the case study. The prediction accuracy of the proposed method is compared with methods in current literature and the analysis results show that the proposed method outperforms all existing methods.

准确预测实时机场容量（即机场验收率（AAR））是实现高效空中交通流量管理的关键。AAR取决于所选的跑道配置，并且两者都受天气条件的影响。尽管已经进行了有关AAR或跑道配置或两者的预测的研究，但预测准确性相对较低，仅考虑单个机场。这项研究提出了一种数据驱动的深度学习框架，用于预测跑道配置和AAR，以支持复杂的多机场系统的有效空中交通管理。这项工作的两个主要贡献是：1）提出的模型对终端空域使用了组合式网格天气预报，而不是孤立的基于站点的终端天气预报，2）该模型捕获了参数学习过程中固有的操作相互依赖性，从而使所提出的建模框架可以同时以更高的精度预测跑道配置和AAR。以纽约大都会区的三个主要机场为例，通过数值实验证明了该方法的有效性。将该方法的预测精度与现有文献中的方法进行了比较，分析结果表明，该方法优于现有方法。以纽约大都会区的三个主要机场为例，通过数值实验证明了该方法的有效性。将该方法的预测精度与现有文献中的方法进行了比较，分析结果表明，该方法优于所有现有方法。以纽约大都会区的三个主要机场为例，通过数值实验证明了该方法的有效性。将该方法的预测精度与现有文献中的方法进行了比较，分析结果表明，该方法优于现有方法。