摘要
为提高多机场旅客吞吐量预测的准确性,简化预测流程,将“手肘法”与高斯混合模型(GMM)系统聚类法结合对多机场系统中的个体进行细化分类。采用面板数据模型分析各变量对不同规模、不同定位的机场旅客吞吐量影响程度,针对不同类型机场建立相应的面板数据吞吐量预测模型,再根据后疫情时期民航业发展现状建立疫情影响因子修正模型,并对预测结果进行修正。以江苏省9个机场的旅客吞吐量分类预测为例,预测结果显示:GMM算法的机场分类效果评价指标CH值为98.732、轮廓系数为0.671 5,较K-means算法分别提高8.3%、69.5%;DB值为0.998 1,较K-means算法降低7%,即GMM算法所得聚类簇内样本间距更小、分类效果更优。模型对9个机场的旅客吞吐量预测误差均介于1.58%~3.95%之间,预测误差波动小、精度较高,具有良好的拟合效果,可用于多机场客流量同步预测。
To improve the accuracy of airport passenger throughput forecasting and simplify the prediction process, this paper combines “elbow method” and Gaussian Mixture Model(GMM) first to classify individuals in multi-airport system.Then selects proper variables base on analysis of relationship between variables and airports with different size and position, so as to establish corresponding panel data model, according to current situation of civil aviation in post-epidemic period, a modified forecasting model which take consideration of epidemic impact is constructed.Taking passenger throughput prediction of 9 airports in Jiangsu as an example, the results show that CH and Silhouette Coefficient, as evaluation index of GMM algorithm, are 98.732 and 0.671 5, improved by 8.3% and 69.5% respectively compared with k-means algorithm, and DB value is 0.998 1, 7% lower than that of k-means, which represent smaller distance of samples in cluster and also a more accurate classification.Nine prediction errors for airports in Jiangsu are between 1.58% and 3.95%, which show smaller error fluctuation, higher accuracy and quite a good imitative effect of forecasting model, indicating that the method can be used for synchronous prediction of passenger flow in multi-airport system.
作者
摆倩倩
李志
BAI Qianqian;LI Zhi(Eastern Airports Group Co.,Ltd.,Nanjing 211100,China;Jiangsu Sub-bureau of East China Regional Air Traffic Management Bureau of Civil Aviation of China,Nanjing 211100,China)
出处
《交通科技与经济》
2022年第6期9-15,共7页
Technology & Economy in Areas of Communications
基金
国家自然科学基金与民航基金联合项目(U1933119)。
关键词
多机场系统
旅客吞吐量预测
后疫情时期
面板数据
高斯混合模型
multi-airport system
passenger throughput prediction
post-epidemic period
panel data
Gaussian mixture model
