基于多源数据和融合网络的铁路客流需求预测研究  

Research on Railway Passenger Demand Forecasting Based on Multi-source Data and Fusion Network

作　　者：刘健 汪颖 Liu Jian;Wang Ying(China Railway Lanzhou Group Co.,Ltd.,Lanzhou,China)

机构地区：[1]中国铁路兰州局集团有限公司,甘肃兰州

出　　处：《科学技术创新》2025年第9期65-68,共4页Scientific and Technological Innovation

摘　　要：本文提出了一种名为VSFNet的多变量时间序列预测模型,用于提高铁路客流量预测的准确性。VSFNet结合特征选择和深度学习,将变量分为高相关性和低相关性两类。高相关性变量通过Transformer提取特征,低相关性变量经PCA降维后,用BiLSTM结合Attention机制提取特征。这些特征在多层结构中融合,并通过跳连接和残差连接增强信息传递。实验结果显示,VSFNet在MSE、MAE和MAPE等指标上优于传统和现代模型,显示出在复杂时序数据处理上的优势。Accurately This paper introduces a model named VSFNet for multivariate time series prediction to enhance the accuracy of railway passenger flow forecasts.VSFNet combines feature selection and deep learning,categorizing variables into high and low correlation groups.High-correlation variables use Transformer for feature extraction,while low-correlation ones are reduced via PCA and processed with BiLSTM and an Attention mechanism.These features are integrated through a multi-layer structure with skip connections and residual links to improve information flow.Experimental results show that VSFNet outperforms traditional and modern models,demonstrating its advantage in handling complex time-series data.

关 键 词：铁路客流预测 多源数据 特征提取 深度学习模型 

分 类 号：U293[交通运输工程—交通运输规划与管理]