A Method for Predicting Speed of Vehicles on Expressways Based on a ST-GCAN Model
-
摘要: 车辆速度是影响高速公路通行效率和安全的重要指标，因此实现对高速公路车辆速度的精准预测有助于减少交通事故进而提升交通智能管控服务水平。基于现有深度学习模型，研究了融合图卷积网络（convo-lutional neural network，GCN）、长短期记忆网络（long short-term memory network，LSTM）和注意力机制的车辆速度预测模型（ST-GCAN）：利用图卷积网络提取复杂高速路网的空间关联特征；使用长短期记忆网络提取车辆速度的历史数据间的时间关联特征；结合注意力机制聚集并分析车辆速度的历史数据和预测值之间的相关性。此外为保障预测模型网络信息完整并解决训练时协变量偏移问题，模型使用密集连接和层归一化技术以提升模型性能表现。利用青海省西宁市的高速公路车辆速度数据集开展实例分析，研究区域包括8个收费站共49条路段，时间跨度为2020年5月1日—8月31日，以小时为步长，共计94 777条数据。实验得到未来1小时高速公路车辆速度的预测效果：平均绝对误差（mean absolute error，MAE）为12.762，均方根误差（root mean square error，RMSE）为21.535，决定系数（R2）为0.651。与传统的时间序列模型和自回归移动平均模型相比，ST-GCAN模型的MAE误差分别降低了约11.1%和19.7%，而对比现有多种深度学习预测模型，ST-GCAN模型的MAE误差降低了约8.0%~10%。ST-GCAN模型在高速公路路网可以实现良好的车辆速度预测效果，满足交通智能管控中的实际预测需求。Abstract: The speed of vehicles on expressways is a significant indicator for describing the effectiveness and safety of road transportation system. Accurate prediction of vehicle speed on expressways can contribute to reduction of traffic accidents and improvement of the level of services. In this sense, a prediction method for vehicle speed, called ST-GCAN, is developed, which integrates graph convolutional neural network (GCN), long short-term memo-ry network (LSTM) and attention mechanism into one model. Graph convolutional network is used to extract the spatial correlations of complex networks of expressways, long-short term memory network is used to extract the temporal correlations of historical data of vehicle speed, and attention mechanism is used to aggregate and analyze the correlation between historical data and predicted vehicle speed. In addition, the model employs dense connec-tions and layer normalization to ensure the integrity of information in the prediction model and to solve the problem of covariate shift during training. The model is tested with a dataset of vehicle speed on expressways of the City of Xining, Province of Qinghai, which contains a total of 94 777 hourly observations on 49 road sections at 8 toll sta-tions from May 1 to August 31, 2020. The ST-GCAN model predicts the vehicle speed in the next hour withthe mean absolute error (MAE) of 12.762%, root mean square error (RMSE) of 21.535%, and R2 of 0.651.Compared to the HA model and the ARIMA model, the MAE of the ST-GCAN model is reduced by 11.1% and 19.7%, respec-tively. Compared to other deep learning models, it is reduced by approximately 8.0% to 10%. In conclusion, the ST-GCAN model can accurately estimate vehicle speed on expressways and shall be able to meet the requirements of intelligent traffic control systems.
-
Key words:
- traffic engineering /
- expressway vehicle speed /
- prediction model /
- neural network /
- attention mechanism
-
表 1 预测指标对比
Table 1. Comparison of forecasting indicators
模型 MAE/（km/h） RMSE/（km/h） R2 HA 14.360 22.867 0.427 ARIMA 15.896 23.137 0.269 SVM 14.342 23.095 0.603 Bi-LSTM 13.912 23.714 0.583 FI-RNNs 13.760 23.721 0.583 HyperNet 13.875 23.714 0.583 Multi-view NN 14.205 22.274 0.632 ST-GCAN 12.762 21.535 0.651 
下载: 导出CSV
-
[1] 刘静, 关伟. 交通流预测方法综述[J]. 公路交通科技, 2004(3): 82-85. https://www.cnki.com.cn/Article/CJFDTOTAL-GLJK200403022.htmLIU J, GUAN W. A summary of traffic flow forecasting meth-ods[J]. Journal of Highway Transportation Research Develop-ment, 2004(3): 82-85. (in Chinese) https://www.cnki.com.cn/Article/CJFDTOTAL-GLJK200403022.htm [2] 袁健, 范炳全. 交通流短时预测研究进展[J]. 城市交通, 2012, 10(6): 73-79. https://www.cnki.com.cn/Article/CJFDTOTAL-CSJT201206015.htmYUAN J, FAN B Q. Synthesis of short-term traffic flow fore-casting research progress[J]. Urban Transport of China, 2012, 10(6): 73-79. (in Chinese) https://www.cnki.com.cn/Article/CJFDTOTAL-CSJT201206015.htm [3] VLAHOGIANNI E I, GOLIAS J C, KARLAFTIS M G. Short term traffic forecasting: Overview of objectives and meth-ods[J]. Transport Reviews, 2004, 24(5): 533-557. doi: 10.1080/0144164042000195072 [4] VAN LINT H, VAN HINSBERGEN C. Short-term traffic and travel time prediction models[J]. Artificial Intelligence Appli-cations to Critical Transportation Issues, 2012, 22(1): 22-41. [5] 叶可江, 田科烺, 须成忠. 1种交通流量的预测方法、系统及终端设备: 201911279828. 4[P]. 2019-12-11.YE K J, TIAN K L, XU C Z. A method, system and terminal equipment of traffic flow prediction: 201911279828. 4[P]. 2019-12-11. (in Chinese) [6] SILVER D, SCHRITTWIESER J, SIMONYAN K, et al. Mas-tering the game of go without human knowledge[J]. Nature, 2017, 550(7676): 354-359. doi: 10.1038/nature24270 [7] PARK D, RILETT L R. Forecasting freeway link travel times with a multilayer feedforward neural network[J]. Comput-er-Aided Civil and Infrastructure Engineering, 2010, 10(5): 357-367. [8] HUANG W, SONG G, HONG H, et al. Deep architecture for traffic flow prediction: Deep belief networks with multitask learning[J]. IEEE Transactions on Intelligent Transportation Systems, 2014, 15(5): 2191-2201. doi: 10.1109/TITS.2014.2311123 [9] LYU Y, DUAN Y, KANG W, et al. Traffic flow prediction with big data: A deep learning approach[J]. IEEE Transac-tions on Intelligent Transportation Systems, 2015, 16(2): 865-873. [10] 陈钰, 张安勤, 许春晖. 基于时空依赖性和注意力机制的交通速度预测[J]. 计算机系统应用, 2021, 30(1): 200-206. https://www.cnki.com.cn/Article/CJFDTOTAL-XTYY202101030.htmCHEN Y, ZHANG A Q, XU C H, Traffic speed prediction based on spatial-temporal dependency and attention mecha-nism[J]. Computer Systems & Applications, 2021, 30(1): 200-206(in Chinese) https://www.cnki.com.cn/Article/CJFDTOTAL-XTYY202101030.htm [11] CAO X, ZHONG Y, ZHOU Y, et al. Inter-active temporal re-current convolution network for traffic prediction in data cen-ters[J]. IEEE Access, 2017 (99): 1-11. [12] KE J, ZHENG H, YANG H, et al. Short-term forecasting of passenger demand under on-demand ride services: A spa-tio-temporal deep learning approach[J]. Transportation Re-search Part C: Emerging Technologies, 2017(85): 591-608. [13] YU H, WU Z, WANG S, et al. Spatiotemporal recurrent con-volutional networks for traffic prediction in transportation networks[J]. Sensors, 2017, 17(7): 1501. doi: 10.3390/s17071501 [14] 陈孟, 干可, 李凯, 等. 基于实时多模态时空数据的时空图卷积网络精准鲁棒交通流预测模型[J]. 公路交通科技, 2021, 38(8): 134-139, 158. https://www.cnki.com.cn/Article/CJFDTOTAL-GLJK202108018.htmCHEN M, GAN K, LI K, et al. A spatial-temporal graph con-volutional network model for accurate and robust traffic flow prediction based on real-time multimodal spatial-temporal data[J]. Journal of Highway and Transportation Research and Development, 2021, 38(8): 134-139, 158. (in Chinese) https://www.cnki.com.cn/Article/CJFDTOTAL-GLJK202108018.htm [15] 张文松, 姚荣涵. 基于时空特性和组合深度学习的交通流参数估计[J]. 交通运输系统工程与信息, 2021, 21(1): 82-89. https://www.cnki.com.cn/Article/CJFDTOTAL-YSXT202101014.htmZHANG W S, YAO R H. Traffic flow parameters estimation based on spatio-temporal characteristics and hybrid deep learning[J]. Journal of Transportation Systems Engineering and Information Technology, 2021, 21(1): 82-89. (in Chinese) https://www.cnki.com.cn/Article/CJFDTOTAL-YSXT202101014.htm [16] 徐先峰, 夏振, 赵龙龙. 基于组合模型的短时交通流预测方法[J]. 测控技术, 2021(3): 117-122. https://www.cnki.com.cn/Article/CJFDTOTAL-IKJS202103022.htmXU X F, XIA Z, ZHAO L L. Short-term traffic flow predic-tion based on combined models[J]. Measurement & Control Technology, 2021(3): 117-122. (in Chinese) https://www.cnki.com.cn/Article/CJFDTOTAL-IKJS202103022.htm [17] 郭芳. 基于轻量级深度学习框架的IP骨干网络流量实时预测研究[D]. 南京: 南京邮电大学, 2020.GUO F, Research on real-time traffic prediction of ip back-bone networks based on light-weighted deep learning frame-work[D]. Nanjing: Nanjing University of Posts, 2020. (in Chinese) [18] 靳嘉曦, 牛文广, 陈炜青, 等. 1种交通数据预测方法、装置及交通工具控制方法: 201911114492. 6[P]. 2019-11-14.JIN J X, NIU W G, CHEN W Q, et al. A traffic data predic-tion method, device and vehicle control method: 201911114492. 6[P]. 2019-11-14. (in Chinese) [19] 张浪文, 张旭, 谢巍, 等. 1种基于滑动窗口长短时记忆网络的交通流量预测方法: 202110326489. 1[P]. 2021-03-26.ZHANG L W, ZHANG X, XIE W, et al. A traffic flow pre-diction method based on sliding window long-term and short-term memory network: 202110326489. 1[P]. 2021-03-26. (in Chinese) [20] 贾兴利, 李双庆, 杨宏志, 等. 基于ATT-LSTM模型的高速公路交通事件持续时长预测[J]. 交通信息与安全, 2022, 40(5): 61-69. doi: 10.3963/j.jssn.1674-4861.2022.05.007JIA X L, LI S Q, YANG H Z, et al. Prediction of the Dura-tion of Freeway Traffic Incidents Based on an ATT-LSTM Mode[J]. Journal of Transport Information and Safety, 2022, 40(5): 61-69(in Chinese) doi: 10.3963/j.jssn.1674-4861.2022.05.007 -
点击查看大图
图(3) / 表(1)
计量
- 文章访问数: 770
- HTML全文浏览量: 293
- PDF下载量: 22
- 被引次数: 0