Automatic Prediction of Traffic Flow Based on Deep Residual Networks
Traffic flow often contains massive amounts of information that is related to location and shows some regularity. And the traffic flow analysis based on trajectory data has become one of the most popular research topics in recent years. With the wide application of deep learning and for its higher accuracy than other approaches, methods such as convolution neural network and deep residual network have been introduced in traffic flow research and achieve good results. However, these methods usually require the training of a large number of parameters, which leads to some problems. For example, frequent manual adjustment is needed, and some parameters cannot be dynamically adjusted with the training process. We find that learning rate plays a crucial role in all parameters, which has important influence on the training speed of the residual network. In other words, the soundness of traffic flow predication results depends on the learning rate. Hence, we propose G4 algorithm to automatically determine the learning rate. It can be adjusted automatically in the process of trajectory data mining, and therefore solve the traffic flow prediction problem. Experiments on real data sets show that our method is effective and superior over some traditional optimizing methods of traffic flow analysis.
KeywordsAutomatic prediction Traffic flow Trajectory Deep residual network Fourier series
- 1.Araki, M., Kanamori, R., Gong, L., Morikawa, T.: Impacts of seasonal factors on travel behavior: basic analysis of GPS trajectory data for 8 months. In: Sawatani, Y., Spohrer, J., Kwan, S., Takenaka, T. (eds.) Serviceology for Smart Service System, pp. 377–384. Springer, Tokyo (2017). https://doi.org/10.1007/978-4-431-56074-6_41CrossRefGoogle Scholar
- 3.Hoermann, S., Bach, M., Dietmayer, K.: Dynamic Occupancy Grid Prediction for Urban Autonomous Driving: A Deep Learning Approach with Fully Automatic Labeling. arXiv preprint arXiv:1705.08781 (2017)
- 5.Zhang, J., Zheng, Y., Qi, D.: Deep spatio-temporal residual networks for citywide crowd flows prediction. In: AAAI (2017)Google Scholar
- 7.Sun, L., et al.: 3DOF Pedestrian Trajectory Prediction Learned from Long-Term Autonomous Mobile Robot Deployment Data. arXiv preprint arXiv:1710.00126 (2017)
- 8.Tong, Y., et al.: The simpler the better: a unified approach to predicting original taxi demands based on large-scale online platforms. In: Proceedings of the 23rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. ACM (2017)Google Scholar