考虑线网结构的LightGBM轨道交通短时客流预测模型

LightGBM Prediction Model of Short-Term Passenger Flow for Rail Transit Considering Network Structure

考虑空间维度特征对利用监督学习预测轨道交通短时客流量的影响,提出结合复杂网络与机器学习理论挖掘车站层面的客流分布规律。通过对原始数据分析,实现对线网结构特征、时间维度特征及数据集的构建,建立基于LightGBM算法的轨道交通短时客流预测模型,并对模型参数进行标定,采用模型评估和特征重要性分析等方法,对模型结果进行分析,对比LightGBM预测模型与XGBoost、随机森林、CatBoost和MLP模型的预测效果。结果表明:考虑线网结构下的LightGBM模型在评估指标MAE,MAPE上表现最优,MAPE最小为13.65%,训练速度较其他模型最多提升至25倍,表现出较强的预测性能。

Considering the influence of spatial features on the short-term passenger flow prediction based on supervised learning for rail transit, this paper proposed a method combining the complex network and the machine learning theory to uncover the distribution law of the passenger flow at the station level. Through the analysis of original data, the network structure features, temporal features, and the final dataset were built, and the short-term passenger flow prediction model for rail transit was constructed based on the LightGBM algorithm. After the parameters were calibrated, the results of the model were analyzed with the help of model evaluation and feature importance analysis. The LightGBM prediction model was compared with XGBoost, Random Forest, CatBoost, and MLP models. The results show that the LightGBM model considering the network structure has the best performance in the evaluation indexes of MAE and MAPE with a minimum MAPE of 13.65%, and the training speed of this model is 25 times higher than other models at most, which shows its strong prediction performance.