期刊
KNOWLEDGE-BASED SYSTEMS
卷 283, 期 -, 页码 -出版社
ELSEVIER
DOI: 10.1016/j.knosys.2023.111184
关键词
Generative model; Network-wide travel time reliability; Prediction of travel time distribution; Automatic vehicle identification data
This paper introduces the travel time reliability-generative adversarial network (TTRGAN) model for predicting network-wide travel time reliability using automatic vehicle identification data. The TTR-GAN model is capable of generating predicted travel time samples without assuming a specific travel time distribution. Experimental results demonstrate that the TTR-GAN model outperforms several benchmark models in terms of statistical, buffer time, and probability distribution measures.
This paper introduces a generative model named the travel time reliability-generative adversarial network (TTRGAN) model for predicting network-wide TTR using automatic vehicle identification data. The TTR-GAN model is capable of generating predicted travel time samples, enabling the assessment of network-wide TTR without the need to assume a specific travel time distribution. In the TTR-GAN model, a combination of graph convolutional networks and long short-term memory (LSTM) neural networks is employed within the GAN framework. When training the TTR-GAN model, special attention is given to adjusting the mean and standard deviation of the generated samples, aiming for a closer resemblance to real samples. Experiments conducted on a road network in China demonstrate the predictive capability of the proposed TTR-GAN model, surpassing several benchmark models such as the LSTM neural network, moving average model, and GAN model in terms of statistical, buffer time, and probability distribution measures. By incorporating the mean and standard deviation into the loss function, the TTR-GAN model achieves an 18.2% reduction in Jensen-Shannon divergence between predicted and real samples. Furthermore, the model's performance in real-world applications is illustrated through a sensitivity test.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据