摘要
乘客出行需求量预测对于优化交通资源配置、改善出行体验、提升交通系统效率具有重要意义。研究旨在利用机器学习方法对城市交通中乘客出行需求量进行预测,通过连续采集21 d的出行需求量数据,构建传统的时间序列模型SARIMAX及多种神经网络模型(包括单向LSTM、双向LSTM、单向LSTM-Attention、双向LSTM-Attention)进行预测,通过MAPE、R^(2)、MAE和RMSE等指标对模型进行评估。结果表明双向LSTM-Attention模型在预测性能上表现最佳,模型的R^(2)为0.8727,RMSE为8.0027,MAPE为0.3343,MAE为5.2952,SARIMAX模型效果最差,引入注意力机制能够有效提升模型的预测精度。研究为城市交通需求预测提供了新的方法和思路,有望在智慧交通规划和交通管理中发挥重要作用。
The passenger travel demand forecast is of great significance for optimizing the allocation of transportation resources,improving the travel experience,and enhancing the efficiency of the transportation system.This paper used machine learning methods to predict passenger travel demands in urban transportation by collecting travel demand data for 21 consecutive days and constructing the conventional time-series model SARIMAX and various neural network models(including uni-directional LSTM,bi-directional LSTM,uni-directional LSTM-Attention,bi-directional LSTM-Attention),and evaluated the models with indexes such as MAPE,R^(2),MAE,and RMSE.The results show that the bi-directional LSTM-Attention model has the optimum forecasting performance,with the R^(2)of 0.8727,RMSE of 8.0027,MAPE of 0.3343,and MAE of 5.2952.The SARIMAX model has the worst forecasting effect,and introducing the attention mechanism can effectively improve the model's forecast accuracy.This study provides new methods and ideas for urban transportation demand forecast,which is expected to play an important role in intelligent transportation planning and traffic management.
作者
杨珵之
YANG Chengzhi(Shanxi Transportation Holding New Energy Development Co.,Ltd.,Taiyuan,Shanxi 030012,China)
出处
《山西交通科技》
2025年第4期147-150,161,共5页
Shanxi Science & Technology of Transportation
