科研之友
微信
新浪微博
Facebook
分享链接
摘要
A short-term traffic flow prediction framework is proposed for urban road networks based on data-driven methods that mainly include two modules. The first module contains a set of algorithms to process traffic flow data. After analysis and repair, a complete data set without outliers is provided as well as a data set containing pairs of road segments that are the most similar to each other in regard to their trends. The second module focuses on multiple time-step short-term forecasting. With a good understanding of the periodicity and randomness of traffic flow, the time series is first decomposed into a trend series and residual series. After reconstructing the two time series, model training and prediction based on a long short-term memory-recurrent neural network (LSTM-RNN) are performed. Finally, the two results are combined together to form the final prediction. A model evaluation is performed using two urban road networks. The results show that the data processing module can effectively improve the data quality, reduce the training time and enhance the model robustness. The LSTM-RNN correctly identifies the time trend and spatial similarity of traffic flow and obtains a more accurate multiple time-step prediction. The proposed framework outperforms other deep learning algorithms and has better accuracy and stability.
- 出版日期2019-4
- 单位华南理工大学
