Comparison of railway passenger flow forecast methods based on ARIMA and LSTM
-
摘要: 精准的客流预测是车站客运组织优化的基础，是提高运营安全和运输效率的有效途径。以江门东站全年进站客流数据为研究对象，分别构建ARIMA时间序列模型与LSTM神经网络模型，从预测精度、计算速度、误差指标评价、模型适应性等方面分析比较两种预测模型对客流预测结果的差异性。结果表明，LSTM模型预测精度和拟合精确度更优，ARIMA模型计算速度更快。研究结果对客流预测方法选择有借鉴意义。Abstract: Accurate passenger flow forcast is the basis of the optimization of passenger transportation organization and an effective way to improve operational safety and transportation efficiency. Taking the annual passenger flow data of Jiangmen East Railway Station as research object, and the ARIMA time series model and the LSTM neural network model were constructed respectively. From the perspectives of forecast accuracy, calculation speed, error index evaluation, and model adaptability, the differences in passenger flow forecast results of two forecast models were analyzed. The results show that the LSTM model has better forecast accuracy and fitting accuracy, and the ARIMA model has faster calculation speed. This study result can provide significant practical implications for the selection of passenger flow forecast methods.
-
Key words:
- railway station /
- ARIMA model /
- LSTM model /
- passenger flow forecast /
- comparative analysis
-
图 3 全年进站客流序列差分后时序图
Figure 3. Time sequence diagram of annual inbound passenger flow sequence after difference
图 4 进站客流差分后的自相关系数图
Figure 4. Autocorrelation coefficient diagram of differential inbound passenger flow
图 5 进站客流差分后的偏自相关系数图
Figure 5. Partial autocorrelation coefficient diagram of differential inbound passenger flow
图 6 ARIMA模型12月份进站客流量预测结果对比图
Figure 6. Comparison of ARIMA model inbound passenger flow forecast results in December
图 7 LSTM模型预测12月份客流量预测结果对比图
Figure 7. LSTM model forecast passenger flow forecast results in December comparison
图 8 LSTM模型12月份进站客流量训练集预测结果对比图
Figure 8. Comparison of the forecast results of LSTM model inbound passenger flow training set in December
图 9 ARIMA模型12月后半月预测结果对比图
Figure 9. Comparison of ARIMA model forecast results in the second half of December
图 10 LSTM模型12月后半月预测结果对比图
Figure 10. Comparison of LSTM model forecast results in the second half of December
表 1 两种模型误差指标计算结果
Table 1. Calculation results of error indexes of two models
预测模型 测试集$ {R}^{2} $ 测试集MAE 测试集MBE ARIMA 0.26702 248.8255 − 31.5423 LSTM 0.52803 181.6317 41.7378 
下载: 导出CSV
-
[1] 李洁, 彭其渊, 文超. 基于LSTM深度神经网络的高速铁路短期客流预测研究[J] . 系统工程理论与实践,2021,41(10):2669 − 2682. [2] 刘星委. 基于ARIMA与长短时记忆神经网络的高速公路交通流预测及比较的研究[D] . 成都: 西南交通大学, 2018. [3] 李德奎, 杜书波, 张鹏. 基于ARIMA和LSTM的城市轨道交通延时客流预测方法比较[J] . 青岛理工大学学报,2021,42(4):135 − 142. [4] 徐映梅, 陈尧. 季节ARIMA模型与LSTM神经网络预测的比较[J] . 统计与决策,2021,37(2):46 − 50. [5] 于洋. 考虑拆船补贴政策的干散货船队更新策略研究[D] . 大连: 大连海事大学, 2017. [6] HOCHREITER S, SCHMIDHUBER J. Long short-term memory[J] . Neural Computation,1997,9(8):1735 − 1780 doi: 10.1162/neco.1997.9.8.1735 [7] 贺菁伟, 杨东谕. 长短期记忆模型在低频数据预测中的应用: 以新冠肺炎疫情对北京市社会消费品零售总额的影响测算为例[J] . 统计理论与实践,2021(1):24 − 28. [8] 邓于佳. 基于百度指数的股票波动数据分析及预测[D] . 贵阳: 贵州财经大学, 2021. [9] 潘念然. 基于ARIMA和LSTM的城市轨道交通客流量预测[J] . 科学技术创新,2022(8):165 − 168. [10] 杨丽, 吴雨茜, 王俊丽, 等. 循环神经网络研究综述[J] . 计算机应用,2018,38(S2):1 − 6, 26. -
下载:
点击查看大图
计量
- 文章访问数: 16
- HTML全文浏览量: 8
- PDF下载量: 0
- 被引次数: 0
