Journal of Geo-information Science >
Attention-based Multi-step Short-term Passenger Flow Spatial-temporal Integrated Prediction Model in URT Systems
Received date: 2022-10-24
Revised date: 2023-02-03
Online published: 2023-04-19
Supported by
National Natural Science Foundation of China(72201029)
National Natural Science Foundation of China(71825004)
National Natural Science Foundation of China(72288101)
China Postdoctoral Science Foundation(2022M720392)
Accurate and reliable short-term passenger flow prediction can support operations and decision-making of the URT system from multiple perspectives. In this paper, we propose a URT multi-step short-term passenger flow prediction model at the network level based on a Transformer-based LSTM network, Depth-wise Attention Block, and CNN network, named as Spatial-Temporal Integrated Prediction Model (STIPM). The STIPM comprises three branches. The first branch takes time-series inflow data as input, and a Transformer-based LSTM network is selected to extract the temporal correlations. The second one takes timestep-based OD data as input, and many spatial and temporal features are captured using Depth-wise Attention Blocks. Meanwhile, timestep-based OD data can better include inter-station relations and global information. The third branch takes Point of Interest data (POI) as input and CNN network is utilized for spatiotemporal features extraction, which can also become the bridge between spatial and temporal features. Moreover, the “Multi-input-multi-output Strategy” for multi-step prediction is used to obtain a longer prediction period and more detailed information under a relatively high forecasting accuracy. The STIPM is applied to two large-scale real-world datasets from the URT system, and the obtained prediction results are compared with ten baselines and four variants from itself, in which STIPM model achieves highest prediction accuracy indicated by RMSE, MAE, and WMAPE evaluations, which demonstrates the superiority and robustness of the STIPM.
ZHANG Jinlei , CHEN Yijie , Panchamy Krishnakumari , JIN Guangyin , WANG Chengcheng , YANG Lixing . Attention-based Multi-step Short-term Passenger Flow Spatial-temporal Integrated Prediction Model in URT Systems[J]. Journal of Geo-information Science, 2023 , 25(4) : 698 -713 . DOI: 10.12082/dqxxkx.2023.220817
表1 各模型预测结果精度 (Dataset 1)Tab. 1 Prediction result evaluation for Dataset 1 |
模型 | 单步预测(10 min) | 双步预测(20 min) | 三步预测(30 min) | ||||||
---|---|---|---|---|---|---|---|---|---|
RMSE | MAE | WMAPE/% | RMSE | MAE | WMAPE/% | RMSE | MAE | WMAPE/% | |
SVR | 24.27 | 13.92 | 17.79 | 28.16 | 15.66 | 20.09 | 31.51 | 17.37 | 22.37 |
DCRNN | 29.97 | 16.45 | 20.28 | 31.42 | 16.63 | 20.51 | 32.26 | 16.85 | 20.82 |
CNN | 28.66 | 15.57 | 17.13 | 29.18 | 15.87 | 17.48 | 29.29 | 16.06 | 17.72 |
GCN | 27.85 | 15.54 | 17.15 | 28.99 | 15.88 | 17.45 | 29.90 | 16.39 | 17.97 |
ST-ResNet | 27.83 | 15.65 | 17.17 | 30.86 | 16.73 | 18.09 | 31.54 | 17.34 | 18.63 |
T- GCN | 29.25 | 15.90 | 17.52 | 30.54 | 16.15 | 17.74 | 30.56 | 17.11 | 18.68 |
ST-GCN | 29.36 | 15.93 | 17.45 | 30.05 | 16.90 | 18.36 | 31.57 | 16.91 | 18.44 |
LSTM | 24.50 | 12.90 | 15.93 | 25.81 | 13.08 | 16.19 | 26.45 | 13.36 | 16.53 |
ConvLSTM | 27.35 | 14.66 | 15.97 | 28.03 | 14.70 | 16.19 | 28.61 | 14.92 | 16.40 |
GWN | 23.17 | 12.68 | 15.70 | 23.19 | 12.97 | 15.98 | 23.58 | 13.19 | 16.31 |
STIPM | 21.67 | 11.91 | 14.65 | 22.45 | 12.19 | 14.96 | 22.60 | 12.34 | 15.23 |
表2 各模型预测结果精度 (Dataset 2)Tab. 2 Prediction result evaluation for Dataset 2 |
模型 | 单步预测(10 min) | 双步预测(20 min) | 三步预测(30 min) | ||||||
---|---|---|---|---|---|---|---|---|---|
RMSE | MAE | WMAPE/% | RMSE | MAE | WMAPE/% | RMSE | MAE | WMAPE/% | |
SVR | 23.65 | 12.16 | 19.58 | 27.91 | 13.81 | 22.22 | 30.85 | 15.20 | 24.42 |
DCRNN | 26.04 | 13.43 | 22.36 | 26.52 | 13.76 | 22.73 | 27.64 | 14.20 | 23.14 |
CNN | 21.55 | 11.72 | 18.65 | 22.01 | 12.08 | 19.14 | 23.89 | 12.53 | 19.83 |
GCN | 21.17 | 11.37 | 18.04 | 22.06 | 11.69 | 18.47 | 23.40 | 12.17 | 19.43 |
ST-ResNet | 25.24 | 13.13 | 20.76 | 25.66 | 13.24 | 21.04 | 25.93 | 13.65 | 21.26 |
T- GCN | 21.87 | 11.73 | 18.74 | 22.60 | 11.95 | 19.02 | 23.24 | 12.28 | 19.50 |
ST-GCN | 20.90 | 11.52 | 18.27 | 22.20 | 11.93 | 18.86 | 22.70 | 12.03 | 18.76 |
LSTM | 22.11 | 11.67 | 18.56 | 22.13 | 11.71 | 18.65 | 22.82 | 11.83 | 18.81 |
ConvLSTM | 20.96 | 11.22 | 17.91 | 21.89 | 11.43 | 18.20 | 21.93 | 11.61 | 18.45 |
GWN | 20.11 | 10.63 | 17.79 | 20.18 | 10.97 | 18.35 | 20.56 | 11.05 | 18.51 |
STIPM | 18.10 | 10.26 | 16.80 | 18.67 | 10.37 | 17.21 | 19.48 | 10.79 | 17.88 |
表3 时间步刻度与真实时间对应关系Tab. 3 Relations between timestep-axis and real time |
时间步刻度 | 真实对应时间 | 时间步刻度 | 真实对应时间 | 时间步刻度 | 真实对应时间 |
---|---|---|---|---|---|
28~136 | 周三 (5:50—24:00) | 137~245 | 周四 (5:50—24:00) | 246~354 | 周五 (5:50—24:00) |
355~463 | 周六 (5:50—24:00) | 464~572 | 周日 (5:50—24:00) |
表4 消融实验预测结果精度(Dataset 1)Tab. 4 Prediction result evaluation of variants for Dataset 1 |
模型 | 单步预测(10 min) | 双步预测(20 min) | 三步预测(30 min) | ||||||
---|---|---|---|---|---|---|---|---|---|
RMSE | MAE | WMAPE/% | RMSE | MAE | WMAPE/% | RMSE | MAE | WMAPE/% | |
STIPM | 21.67 | 11.91 | 14.65 | 22.45 | 12.19 | 14.96 | 22.60 | 12.34 | 15.23 |
STIPM-No OD | 22.86 | 12.53 | 15.38 | 23.04 | 12.69 | 15.68 | 24.05 | 12.89 | 15.88 |
STIPM-No POI | 22.84 | 12.41 | 15.28 | 23.72 | 12.47 | 15.42 | 23.91 | 12.72 | 15.59 |
STIPM-LSTM | 22.51 | 12.41 | 15.25 | 23.34 | 12.95 | 15.86 | 24.22 | 0.65 | 16.10 |
STIPM-Acc OD | 23.28 | 12.33 | 15.16 | 23.78 | 12.88 | 15.87 | 24.11 | 12.98 | 16.02 |
[1] |
张春辉, 宋瑞, 孙杨. 基于卡尔曼滤波的公交站点短时客流预测[J]. 交通运输系统工程与信息, 2011, 11(4):154-159.
[
|
[2] |
王莹, 韩宝明, 张琦, 等. 基于SARIMA 模型的北京地铁进站客流量预测[J]. 交通运输系统工程与信息, 2015, 15(6):205-211.
[
|
[3] |
|
[4] |
|
[5] |
张金雷. 城市轨道交通线网短时客流预测方法研究[D]. 北京: 北京交通大学, 2021.
[
|
[6] |
|
[7] |
|
[8] |
|
[9] |
|
[10] |
|
[11] |
陈深进, 薛洋. 基于改进卷积神经网络的短时公交客流预测[J]. 计算机科学, 2019, 46(5):175-184.
[
|
[12] |
|
[13] |
|
[14] |
|
[15] |
王海洋. 城市轨道交通系统突发客流的分析与预测预警研究[D]. 上海: 上海交通大学, 2019.
[
|
[16] |
|
[17] |
|
[18] |
|
[19] |
|
[20] |
|
[21] |
|
/
〈 |
|
〉 |