基于GCN的杭州地铁客流量分析与预测

GCN-Based Passenger Flow Analysis and Prediction for Hangzhou Metro

摘要: 本文利用2021年杭州地铁的支付宝移动支付数据，通过详细的数据预处理和特征分析，构建了一个基于GCN的预测模型，旨在提高对未来一天内不同时间段的客流量预测准确性。该模型特别考虑了地理空间相关性、分布差异性以及动态分布相关性等多个维度，以便更好地理解站点间的关联性和乘客流动模式。实验结果显示，该模型在预测进站和出站客流量上的均方误差分别为2245.45和2127.09，表现优于XGBoost和LightGBM等传统机器学习模型，能为地铁运营者提供更加精准的客流量预测数据，帮助其在国庆假期等特殊时期更好地进行客流管理和资源调度。

Abstract: This study leverages Alipay mobile payment data from Hangzhou Metro in 2021 to construct a GCN-based prediction model through detailed data preprocessing and feature analysis. The aim is to enhance the accuracy of predicting passenger flows across different time periods within a day. The model specifically incorporates multiple dimensions such as geographical spatial correlations, distribution differences, and dynamic distribution correlations to better understand inter-station relationships and passenger movement patterns. Experimental results demonstrate that the model achieves mean squared errors of 2245.45 for entry flows and 2127.09 for exit flows, outperforming traditional machine learning models like XGBoost and LightGBM. This improved accuracy provides metro operators with more precise passenger flow predictions, aiding in effective crowd management and resource allocation during special periods such as National Day holidays.