短期交通流量預測之議題已發展多年，為了改善預測準確度，許多研究嘗試以不同之方法預測，目標為貼近真實資料。相較於傳統統計方法建構模型，深度學習之技術能達到更精準之預測，因此本研究藉由長短期記憶神經網路(Long Short-term Memory Neural Network, LSTM NN)，門控循環單元網路(Gated Recurrent Unit Network, GRU)與時序卷積網路(Temporal Convolutional Networks, TCN)模型預測流量，並以交通部所提供之｢高速公路電子收費交通資料蒐集支援系統」(Traffic Data Collection System, TDCS)數據來實證模型。首先將資料整理成歷史流量之時間序列作為輸入，並調整模型中之超參數(hyper-parameters)讓根均方誤差(Root Mean Squared Error, RMSE)達到收斂， 進而依不同的車種預測未來5分鐘之短期交通流量。結果顯示，此模型與真實資料擬合情況良好，且模型具有同時預測時間及空間之能力，能夠貼近真實交通情況，並由於時序卷積網路模型運算速度大幅優於其他兩個模型，故推論此模型為三者中最佳之預測流量模型。

Studies of short-term traffic flow prediction have been conducted for many years. Many studies with different models and algorithms have been proposed to improve prediction accuracy. Deep learning has proved to have a higher accuracy than traditional statistics analyses. As a result, this study proposes traffic flow prediction models based on long short-term memory neural network (LSTM NN), Gated Recurrent Unit Network (GRU), and Temporal Convolutional Networks (TCN). The data in this research is derived from Traffic Data Collection System (TDCS) in Taiwan, and the historical time series of traffic flows are adopted as input. Then traffic flow of different types of vehicles for 5 minutes in the future are predicted by proposed models. The empirical results show that the prediction data fit the true data well using all of the proposed models, which could be used to predict spatial-temporal traffic flow from actual traffic conditions. In conclusion, the TCN performs best because of its computational efficiency.

REFERENCES
[1] Bai, S., et al. (2018) An Empirical Evaluation of Generic Convolutional and Recurrent Networks for Sequence Modeling. arXiv e-prints
[2] Bottou, L. J. O.-l. l. i. n. n. (1998). Online learning and stochastic approximations. On-line learning in neural networks, Cambridge University Press 17: 142.
[3] Box, G. E. P. and G. M. Jenkins (1976). Time series analysis : forecasting and control / George E.P. Box and Gwilym M. Jenkins. San Francisco, Holden-Day.
[4] Cho, K., et al. (2014) Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation. arXiv e-prints
[5] Fu, R., et al. (2016). Using LSTM and GRU neural network methods for traffic flow prediction. Chinese Association of Automation (YAC), Youth Academic Annual Conference of, IEEE.
[6] Fukushima, K. and S. Miyake (1982). "Neocognitron: A new algorithm for pattern recognition tolerant of deformations and shifts in position." Pattern Recognition 15(6): 455-469.
[7] Goodfellow, I., et al. (2016). Deep learning, MIT press Cambridge.
[8] He, K., et al. (2016). Deep Residual Learning for Image Recognition. 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).
[9] Hochreiter, S. and J. J. N. c. Schmidhuber (1997). "Long short-term memory." Neural Comput 9(8): 1735-1780.
[10] Karpathy, A., et al. (2014). Large-scale video classification with convolutional neural networks. Proceedings of the IEEE conference on Computer Vision and Pattern Recognition.
[11] Kim, W. K., et al. (2017). "A Study of Deep CNN-Based Classification of Open and Closed Eyes Using a Visible Light Camera Sensor." Sensors 17(7).
[12] LeCun, Y., et al. (2010). Convolutional networks and applications in vision.
Proceedings of 2010 IEEE International Symposium on Circuits and Systems.
[13] Lippi, M., et al. (2013). "Short-Term Traffic Flow Forecasting: An Experimental Comparison of Time-Series Analysis and Supervised Learning." IEEE Transactions on Intelligent Transportation Systems 14(2): 871-882.
[14] Lv, Y., et al. (2015). "Traffic Flow Prediction With Big Data: A Deep Learning Approach." IEEE Transactions on Intelligent Transportation Systems 16(2): 865-873.
[15] Ma, X., et al. (2015). "Long short-term memory neural network for traffic speed prediction using remote microwave sensor data." Transportation Research Part C: Emerging Technologies 54: 187-197.
[16] Ministry-of-Transportation-and-Communications, F. B. (2018). "ETC Operating Statistics." Retrieved 0830, 2018, from https://www.freeway.gov.tw/etc/publish.aspx?NID=1949&P=4933.
[17] Ministry-of-Transportation-and-Communications, F. B. (2018). "Traffic Data Collection System (TDCS)." Retrieved 0701, 2018, from http://tisvcloud.freeway.gov.tw/history/TDCS/.
[18] Polson, N. G. and V. O. Sokolov (2017). "Deep learning for short-term traffic flow prediction." Transportation Research Part C: Emerging Technologies 79: 1-17.
[19] Ramos, C., et al. (2008). "Ambient Intelligence—the Next Step for Artificial Intelligence." IEEE Intelligent Systems 23(2): 15-18.
[20] Scherer, D., et al. (2010). Evaluation of pooling operations in convolutional architectures for object recognition. Artificial Neural Networks–ICANN 2010, Springer: 92-101.
[21] Sivaraman, S. and M. M. Trivedi (2010). "A General Active-Learning Framework for On-Road Vehicle Recognition and Tracking." IEEE Transactions on Intelligent Transportation Systems 11(2): 267-276.
[22] Sutskever, I., et al. (2011). Generating text with recurrent neural networks. Proceedings of the 28th International Conference on Machine Learning (ICML-11).
[23] Szegedy, C., et al. (2015). Going deeper with convolutions. Proceedings of the IEEE conference on computer vision and pattern recognition.
[24] Tian, Y. and L. Pan (2015). Predicting short-term traffic flow by long short-term memory recurrent neural network. Smart City/SocialCom/SustainCom (SmartCity), 2015 IEEE International Conference on, IEEE.
[25] Van Lint, J., et al. (2002). "Freeway travel time prediction with state-space neural networks: modeling state-space dynamics with recurrent neural networks." Transportation Research Record(1811): 30-39.
[26] Vedaldi, A. and K. Lenc (2015). Matconvnet: Convolutional neural networks for matlab. Proceedings of the 23rd ACM international conference on Multimedia, ACM.
[27] Wenger, E. (1987). Artificial intelligence and tutoring systems: computational and cognitive approaches to the communication of knowledge, Morgan Kaufmann.
[28] Williams, B. M. and L. A. J. J. o. t. e. Hoel (2003). "Modeling and forecasting vehicular traffic flow as a seasonal ARIMA process: Theoretical basis and empirical results." Journal of Transportation Engineering 129(6): 664-672.
[29] Wu, Y., et al. (2018). "A hybrid deep learning based traffic flow prediction method and its understanding." Transportation Research Part C: Emerging Technologies 90: 166-180.
[30] Yu, H., et al. (2017). "Spatiotemporal Recurrent Convolutional Networks for Traffic Prediction in Transportation Networks." Sensors 17(7).
[31] Zhang, D. and M. R. Kabuka (2017). Combining Weather Condition Data to Predict Traffic Flow: A GRU Based Deep Learning Approach. 2017 IEEE 15th Intl Conf on Dependable, Autonomic and Secure Computing, 15th Intl Conf on Pervasive Intelligence and Computing, 3rd Intl Conf on Big Data Intelligence and Computing and Cyber Science and Technology Congress(DASC/PiCom/DataCom/CyberSciTech).
[32] Zhang, Y. and G. Huang (2018). "traffic flow prediction model based on deep belief network and genetic algorithm." IET Intelligent Transport Systems 12(6): 533-541.
[33] Zhao, Z., et al. (2017). "LSTM network: a deep learning approach for short-term traffic forecast." IET Intelligent Transport Systems 11(2): 68-75.
[34] Zhou, S., et al. (2016). "Spatial–temporal convolutional neural networks for anomaly detection and localization in crowded scenes." Signal Processing: Image Communication 47: 358-368.