为实现不同场景下的铁路客运量预测，提升预测精度，提出一个通用的中长期铁路客运量预测理论框架，包括数据采集层、特征挖掘层和模型预测层：采集并整合预测所需的多源数据，进行特征构建和挖掘后，将特征输入预测模型进行预测。基于该框架，构建了基于EMD-LightGBM的中长期铁路客运量预测模型。最后，以1981—2021年的铁路客运量数据为例，解析了预测框架的构建过程，并对特征挖掘技术的应用效果进行了分析。结果表明，特征挖掘方法能大幅提升铁路客运量的预测精度和预测效率;所建预测框架可推广应用于线路或地区等不同场景的铁路客运量预测。

In order to predict railway passenger volume in different scenarios and improve the prediction accuracy, a universal theoretical framework for railway passenger volume prediction was proposed, including data collection layer, feature mining layer, and model prediction layer, which means to collect and integrate multi-source data required for prediction, construct and mine features, and input them into the prediction model for prediction. Based on this framework, an EMD-LightGBM medium and long-term railway passenger volume prediction model was constructed. Finally, taking the railway passenger volume data from 1981 to 2021 as an example, the construction process of the prediction framework and the application effect of feature mining technology were analyzed. The results indicated that feature mining methods can greatly improve the prediction accuracy and efficiency of railway passenger volume; the established prediction framework and model can be widely applied to railway passenger volume prediction in different scenarios of routes or regions.