摘要
由于环境因素、节假日及大型活动等外部因素带来的高度非线性和不确定性,使得预测轨道交通网络的多粒度动态客流是一项极具挑战的任务。为此,开发能够应对此类复杂问题的高精度预测模型至关重要。提出一种融合复杂外部因素的轨道交通客流多步预测模型(STEF-GCN模型),该模型深度融合具有强相关性的复杂外部因素,并考虑轨道交通客流数据的时空相关性。研究结果表明,STEF-GCN模型性能显著优于其他基线模型,具体表现为:在短期与长期预测中均能实现高精度,在短期预测中其RMSE值和MAE值分别降低了8.60%~63.80%和10.05%~66.90%;模型能够更好地预测客流峰值与低谷值,把握客流的波动性;对突发情况引发的客流波动具有更强适应能力,且抗干扰性能优异。该模型有望辅助轨道交通运营方制定科学决策,提升服务效率。
Due to the high nonlinearity and uncertainty caused by external factors such as environmental conditions,holidays,and large-scale events,it's significant challenges to predict multi-granularity dynamic passenger flows in rail transit networks so that developing high-precision prediction models that can handle complexities is crucial.A Spatio-Temporal Extermal Factor-integrated Graph Convolutional Network(STEF-GCN)model is proposed,which deeply incorporates strongly correlated complex external factors and considering the spatiotemporal correlation of rail transit passenger flow data.The result shows that the performance of STEF-GCN model significantly is better than the other baseline methods.Specifically manifested as achieving high precision in both short-term and long-term predictions.RMSE and MAE are reduced by 8.60%~63.80%and 10.05%~66.90%,respectively in short-term forecasting;The model can better predict peak and off-peak passenger flow,capturing the fluctuations in passenger traffic;The model exhibits strong adaptability to fluctuations of passenger flow caused by sudden situations and excellent anti-interference performance.The STEF-GCN model is expected to assist rail transit operators in making scientific decisions and improving service efficiency.
作者
赵莉
ZHAO Li(China Academy of Urban Planning and Design,Beijing 100044,China)
出处
《市政技术》
2025年第8期145-151,共7页
Journal of Municipal Technology
基金
北京市自然科学基金项目(8252005)。
关键词
轨道交通
客流预测
复杂外部因素
深度学习
时空特征
rail transit
passenger flow prediction
complex external factors
deep learning
spatiotemporal features
