计算机工程与应用 ›› 2020, Vol. 56 ›› Issue (15): 140-146.DOI: 10.3778/j.issn.1002-8331.1905-0104

• 模式识别与人工智能 • 上一篇    下一篇

基于ST-DCGAN的时序交通流量数据补全

袁瑶瑶,康雁,李浩,牛瑞丞,梁文韬,李晋源   

  1. 云南大学 软件学院,昆明 650500
  • 出版日期:2020-08-01 发布日期:2020-07-30

Timing Traffic Flow Data Completion Based on ST-DCGAN

YUAN Yaoyao, KANG Yan, LI Hao, NIU Ruicheng, LIANG Wentao, LI Jinyuan   

  1. School of Software, Yunnan University, Kunming 650500, China
  • Online:2020-08-01 Published:2020-07-30

摘要:

时序交通流量数据作为一种新型城市数据,对智能交通和智慧城市的发展有着重要的意义,但是由于各种原因使得收集的交通数据存在大量的缺失,因此如何有效地补充缺失流量数据成为目前急需解决的问题。提出的ST-DCGAN模型利用了基于DCGAN网络的思想,引入补全损失函数和判别损失函数作为模型新的目标函数,通过生成器和鉴别器相互博弈的原理学习区域流量数据之间的时空特征性,在常规的缺失数据补全的基础上利用数据生成思想进行了区域时序交通流量数据的补全,从而为交通流量缺失值提出一种新的补全方法。实验以北京TaxiBJ GPS开源数据集为基础,并用RMSE评估函数分析上述算法对缺失交通流量补全的效果,实验结果表明提出的方法与所比较补全方法相比,效果更好。

关键词: ST-DCGAN网络, 时序数据, 缺失交通流量补全

Abstract:

As a new type of time series city data, time series traffic flow data is of great significance to the development of intelligent transportation and smart city. However, due to various reasons, the collected traffic data has a large number of missing, so how to effectively supplement the missing traffic data becomes an urgent problem. The ST-DCGAN model proposed in this paper utilizes the idea of DCGAN network, introduces the complete loss function and the discriminant loss function as the new objective function of the model, and learns the spatio-temporal characteristics between the regional traffic data through the principle of the game between the generator and the discriminator. On the basis of the conventional missing data completion, the data generation idea is used to complete the regional time series traffic flow data, so as to propose a new complement method for the traffic flow missing value. Based on the Beijing TaxiBJ GPS open source dataset, the RMSE evaluation function is used to analyze the effect of the above algorithm on the missing traffic flow completion. The experimental results show that the proposed method is better than compare complementation method.

Key words: ST-DCGAN network, time series, missing traffic flow completion