基于GCN-CNN模型的轨道交通短时客流预测
Short-Term Passenger Flow Prediction in Rail Transit Based on GCN-CNN Model

准确、可靠的短时客流预测对于城市轨道交通的运营管理至关重要，能为运营管理者提供优化决策的依据，从而改善乘客服务质量和提升交通运营效率。文章基于数据驱动的方法，对北京地铁连续5周25个工作日的自动售票系统(Automatic Fare Collection，简称AFC)刷卡数据进行了详细的分析。通过对AFC数据的清洗和预处理，提取15 min时间粒度的进站客流时间序列并进行归一化处理。采用图卷积神经网络(Graph Convolutional Network，简称GCN)和二维卷积神经网络(Convolutional Neural Network，简称CNN)的组合模型对客流进行预测，模型输入为周模式、日模式、实时模式三个模式下的短时进站客流序列，综合考虑不同时间尺度上的客流变化。为验证模型的有效性和预测的精度，选用真实的北京地铁客流数据集进行实例分析，并运用均方根误差、决定系数、平均绝对误差、加权平均绝对百分误差等评估指标评估客流预测精度。结果表明，与传统的单一模型相比，GCN-CNN组合模型的准确性和精度均取得了显著的提高。
Accurate and reliable short-term passenger flow prediction is crucial for the operation and management of urban rail transit, providing the basis for operational managers to optimize decision-making, thereby improving passenger service quality and enhancing transport efficiency. This paper presents a detailed analysis of card swipe data from the Automatic Fare Collection (AFC) system of the Beijing Metro for 25 consecutive working days over 5 weeks. After cleaning and preprocessing the AFC data, time series of inbound passenger flow with 15-minute granularity is extracted and normalized. The combined model of Graph Convolutional Network (GCN) and 2D Convolutional Neural Network (CNN) is utilized for passenger flow prediction, with inputs including short-term inbound passenger flow sequences in weekly, daily, and real-time modes, comprehensively considering passenger flow changes at different time scales. To verify the validity and accuracy of the model, a real Beijing metro passenger flow dataset is analyzed, utilizing evaluation metrics such as root mean square error, coefficient of determination, average absolute error, and weighted average absolute percentage error. The results demonstrate significant improvements in accuracy and precision compared to traditional single-model approaches.