Page 185 - 《软件学报》2021年第12期

于瑞云 等:基于可变形卷积时空网络的乘车需求预测模型 3849

实验最终采用此参数.
本文研究发现,时间步长 h 是进行时间特征提取的重要参数.为了确定其对模型性能的影响,所以本文在
CD2Data 数据集下对 h 的不同取值进行对比实验.具体实验结果见表 2.

Table 2 Performance comparison of different sequence length
表 2 不同时间步长性能对比
MAPE RMSE
h=2 0.242 9 16.83
h=4 0.225 7 14.57
h=6 0.209 1 13.42
h=8 0.207 7 12.42
h=10 0.208 3 12.59
h=12 0.212 1 13.12

由表 2 可以得出,模型误差整体上随时间步长 h 的增长而下降.由此可以说明时间步长对时间特征提取的
重要性.当时间步长为 8 时,模型性能最佳.此外,h 增加到 8 以上时模型性能略有下降.一个潜在的原因是:在考虑
更长的时间依赖性时,需要学习更多的参数,进而使训练变得更加困难.

4 总 结

本文在乘车需求预测场景下对传统时空模型进行改进,提出多组件的可变形卷积时空网络 DCSN.本模型
将传统时空模型中的卷积神经网络改进为可变形卷积神经网络,使对目标区域的空间特征提取效果得到提升.
本文还将 POI 对乘车需求的影响因素加入到模型中,弥补了局部 DCN 模型在空间上的距离局限性,使 DCSN 模
型效果优于其他已有的乘车需求预测方法.
未来,本文将在两个主要方面对模型进行改进.将更多对乘车需求有显著影响的外部因素进行更合适的建
模,并将其嵌入到模型中;本文在进行预测时仅考虑了近 4 个小时的历史需求对未来需求的影响,在接下来的研
究中,本文将进一步考虑长周期(如日或周)历史数据对未来需求的影响并将其加入到模型中,使预测结果更加
准确.

致谢 实验数据来自滴滴出行“盖亚”数据开放计划.

References:
[1] Williams BM, Hoel LA. Modeling and forecasting vehicular traffic flow as a seasonal ARIMA process: Theoretical basis and
empirical results. Journal of Transportation Engineering, 2003,129(6):664−672.
[2] Li X, Pan G, Wu Z, et al. Prediction of urban human mobility using large-scale taxi traces and its applications. Frontiers of
Computer Science, 2012,6(1):111−121.
[3] Moreira-Matias L, Gama J, Ferreira M, et al. Predicting taxi–passenger demand using streaming data. IEEE Trans. on Intelligent
Transportation Systems, 2013,14(3):1393−1402.
[4] Abadi A, Rajabioun T, Ioannou PA. Traffic flow prediction for road transportation networks with limited traffic data. IEEE Trans.
on Intelligent Transportation Systems, 2014,16(2):653−662.
[5] Wu F, Wang H, Li Z. Interpreting traffic dynamics using ubiquitous urban data. In: Proc. of the 24th ACM SIGSPATIAL Int’l Conf.
on Advances in Geographic Information Systems. 2016.
[6] Tong Y, Chen Y, Zhou Z, et al. The simpler the better: A unified approach to predicting original taxi demands based on large-scale
online platforms. In: Proc. of the the 23rd ACM SIGKDD Int’l Conf. 2017.
[7] Yi H, Jung HJ, Bae S. Deep neural networks for traffic flow prediction. In: Proc. of the IEEE Int’l Conf. on Big Data & Smart
Computing. 2017. 328−331.
[8] Wang D, Cao W, Li J, et al. DeepSD: Supply-demand prediction for online car-hailing services using deep neural networks. In:
Proc. of the IEEE 33rd Int’l Conf. on Data Engineering (ICDE). 2017. 243−254.
[9] Zhao K, Khryashchev D, Freire J, et al. Predicting taxi demand at high spatial resolution: Approaching the limit of predictability. In:
Proc. of the IEEE Int’l Conf. on Big Data. 2016. 833−842.