中文摘要：

为了有效利用机场容量资源,克服现有随机GHP模型中容量预测存在的人为误差,研究了机场容量混合聚类算法。将每天的容量按照30 min间隔划分为多个区间,每个区间对应着1个容量值,这样每天的容量就作为1个容量样本。采集国内某机场半年的容量样本,采用k-means和SOM神经网络的混合聚类算法,确定机场典型容量样本,计算相应的概率,建立典型容量样本树,并应用于随机GHP的静态和动态模型。仿真结果表明：与不执行GHP相比,静态和动态模型的总延误损失分别减少了32.7%和52.7%,验证了混合聚类算法的可行性以及典型容量样本树的实用性。

英文摘要：

In order to make full use of airport capacity resources and eliminate existing human prediction error in stochastic ground holding policy（GHP） model,a mixed clustering algorithm was researched.Daily capacity was divided into several intervals in accordance with 30 min,each interval corresponded to a certain capacity value,and the capacity of one day was a capacity scenario.The capacity scenarios of an airport in half a year were collected,typical capacity scenarios were produced by using self-organizing-maps（SOM） neural network and k-means clustering algorithm,and the probability of each capacity scenario was calculated.Typical capacity scenario tree was constructed and applied in stochastic static and dynamic GHP models.Simulation result shows that compared with no-GHP case,the total delay costs of static and dynamic GHP models reduce by 32.7% and 52.7% respectively.So the mixed algorithm is feasible,and the typical capacity scenario tree is practical.1 tab,4 figs,15 refs.