| 徐东伟* **,彭鹏**,何德峰**.数据缺失下的短时交通流预测[J].高技术通讯(中文),2021,31(9):934~941 |
| 数据缺失下的短时交通流预测 |
| Short-term traffic flow prediction with missing data |
| |
| DOI：10.3772/j.issn.1002-0470.2021.09.004 |
| 中文关键词: 生成对抗网络（GAN）； 门控循环单元（GRU）； 交通数据缺失； 短时交通流预测 |
| 英文关键词: generative adversarial network (GAN), gated recurrent unit (GRU), traffic data missing, short-term traffic flow prediction |
| 基金项目: |
|
| 摘要点击次数: 2011 |
| 全文下载次数: 1634 |
| 中文摘要: |
| 在实际交通数据收集过程中，采集设备故障、维修等问题均易导致采集到的交通数据存在一定的缺失。针对交通数据缺失情况下的交通流预测问题，本文提出了一种基于生成式对抗网络的短时交通流预测模型。该模型由生成网络和判别网络两部分组成。其中，生成网络由全连接层和门控循环单元（GRU）构成，以编码-解码的形式完成对未来交通状态的预测输出；判别网络由多层全连接层构成，通过Wasserstein距离的计算完成对真假样本的有效判断。实验结果表明，本文提出的模型不仅适用于不同比例数据缺失下的短时交通流预测，而且其预测表现优于其他对比模型。 |
| 英文摘要: |
| In the process of actual traffic data collection, problems such as equipment failure and maintenance are likely to cause traffic data missing. Aiming at the problem of traffic flow prediction with missing data, a short-term traffic flow prediction model based on generative adversarial networks is proposed. The model is composed of two parts: generating network and discriminating network. The generation network is composed of the fully connected layer and the gated recurrent unit (GRU), which completes the prediction output of the future traffic flow in the form of encoding-decoding; the discriminant network is composed of multiple fully connected layers, which discriminates the real traffic data and fake data by the compute of Wasserstein distance. The experimental results show that the proposed model is not only suitable for short-term traffic flow prediction with different proportions of data missing, but also has a better prediction performance than other comparative models. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
