首页|基于深层时空图卷积网络的短时到港客流时空分布预测

基于深层时空图卷积网络的短时到港客流时空分布预测

Spatio-temporal distribution prediction of short-term arriving passenger flow based on deep spatio-temporal graph convolutional network

扫码查看
原文链接
  • NETL
  • NSTL
  • 维普
  • 万方数据
针对空港综合交通枢纽各区域短时到港客流感知能力较弱的问题,提出一种基于深层时空图卷积网络的预测方法.以空港综合交通枢纽的空间连通特点和到港旅客的行为规律为依据,构建深层图卷积网络提取临近时间段内到港客流量分布的空间特征,并运用门控循环单元提取空间特征序列的时间依赖性,同时利用当前与历史航班信息对预测结果进行修正,实现对目标时间段内各区域内到港客流的预测.基于国内某大型空港综合交通枢纽内到港客流的历史数据展开验证,结果表明,与代表性的预测模型(历史均值模型、自回归差分滑动平均模型、支持向量机回归模型、长短时记忆神经网络、门控循环单元模型、时间图卷积网络)相比,该方法在测试集上的均方根误差和平均绝对值误差均取得最小值,相较于预测精度第二的时间图卷积网络,预测时间范围为 5、15、30 min时,均方根误差分别降低了 4.19%、7.15%、7.79%,平均绝对值误差分别降低了 9.72%、5.05%、8.89%,说明该方法能够更真实地反映不同区域不同时间段内的客流变化趋势,有助于合理地进行空港综合交通枢纽的运力资源配置.
For the weak perceptual ability of short-term arriving passenger flow in each region of airport comprehensive trans-portation hub,a prediction method based on deep spatio-temporal graph convolutional network is proposed.Ac-cording to the spatial connectivity characteristics of airport comprehensive transportation hub and the behavioral patterns of arriving passenger,a deep graph convolutional network is constructed to extract the spatial characteris-tics of the distribution of arriving passenger flow in adjacent time periods,and gated recurrent unit is applied to ex-tract the temporal dependence of the spatial feature sequence.Moreover,current and historical flight information are used to correct the prediction results and achieve the prediction of arriving passenger flow in each region within the target period.Validation based on historical data of arriving passenger flow within a comprehensive transporta-tion hub of a large domestic airport is conducted,the results show that compared with the representative prediction models(history average model,autoregressive integrated moving average model,support vector regression model,long short-term memory neural network,gated recurrent unit model,temporal-graph convolutional network),this method achieved the minimum value of root mean square error(RMSE)and mean absolute error(MAE)on the test set.Compared with the temporal-graph convolutional network with the second highest prediction accuracy,when the predicting step is 5 min,15 min and 30 min,the RMSE decreased by 4.19%,7.15%,7.79%and the MAE de-creased by 9.72%,5.05%and 8.89%,respectively,indicating that this method can more accurately reflect the trend of passenger flow changes in different regions and time periods,which can help the rational allocation of transporta-tion capacity resources for airport comprehensive transportation hub.

air transportationshort-term passenger flow predictiondeep graph convolutional networkflight information correctiondeep learningairport comprehensive transportation hub

张红颖、贾驰、李彪

展开 >

中国民航大学 电子信息与自动化学院,天津 300300

中国民航大学 航空工程学院,天津 300300

航空运输 短时客流预测 深层图卷积网络 航班信息修正 深度学习 空港综合交通枢纽

国家重点研发计划

2018YFB1601200

2024

中国民航大学学报
中国民航大学

中国民航大学学报

影响因子：0.363
ISSN：1674-5590
年,卷(期)：2024.42(2)
  • 15