即時交通預測在實現綠色運輸的目標中扮演著極為重要的角色。在過去的文獻中，有各種基於以歷史資料計算參數的交通預測模組被提出來，其中像是自回歸滑動平均模型(Autoregressive Moving Average Model)及神經網路模型(Neural Network models)。在這種基於以歷史資料計算參數的交通預測模組發展過程中，需要大量的交通資料及準確的地理資料。因此，假設有一個平台能夠提供開源的地理資料及持續不斷更新的交通資料，整個預測模組的發展時程便能大幅減化，然而，目前在學術上卻沒有看到相關的平台被提出來。在本篇論文中，我們提出一個交通預測實驗平台，除了提供開源的地理資料、定期更新的交通資料也支援使用者可以重新佈置他們自己的最短路徑演算法及預測模組。另一方面，在過去提出的預測模組中，大部分存在著必須使用大量的參數預測，導致複雜度太高而發生過度擬合(over-fit)的問題，使得預測的效能降低。此外，過去的預測方法中，針對交通預測，大部分只考慮了時間(temporal)及空間(spatial)因素與預測值之間的相依性(correlation)，而忽略紅綠燈因素對於交通預測的影響。因此，我們除了發展實驗平台外，也提出了一個同時考慮時間、空間及紅綠燈因素的低複雜度半參數(semi- parametric)預測模組。這個預測模組著重在路徑旅行時間的預估，主要可以分為兩個部分，分別為在紅綠燈前排隊時間的預估模組及道路速率的預測模組。在紅綠燈前排隊時間的預估模組中，我們從政府取得相關的紅綠燈資料，再進一步的對路口車輛行為進行假設；而在道路速率的預測模組中，我們採用過去被提出來的車輛速率預測模組，最後，在整個旅行時間的預估效能分析實驗中，和Google相比，我們預估出來的旅行時間更貼近現實的交通狀況。

Real-time road traffic prediction plays a crucial role in realizing eco-driving for green transportation. In the literature, a variety of data-driven parametric traffic prediction methods, e.g., the ARMA and Neural Network models, has been proposed. In the DDP prediction methods, the prediction model development can be divided into several steps, where traffic data and geographic data are needed. If there is a platform which can provide open source geographic data and continually updated traffic data, the prediction procedure can be simplified. However, there is no such the platform in the literature. In this paper, a traffic prediction experiment platform is proposed, which provides the open source geographic data, periodically updated traffic data and supports users to deploy their own shortest path algorithm and prediction model on it. On the other hand, a common problem of these methods is that their model complexity requires estimating a large amount of parameters. Because of such complexity, these methods may typically over-fit, leading to poor performance for predicting the future data. Moreover, these methods focus on temporal/spatial correlation analysis, but ignore the design of traffic lights model. Because of these drawbacks, in addition to the development of the experiment platform, we proposed a low-complexity semi- parametric prediction model, where the temporal, spatial correlation and traffic light effect are considered at the same time. The proposed prediction model focuses on the routing travel time estimation, which can be divided into two parts: the queuing time for traffic light estimation and the vehicle velocity estimation. For the queueing time analysis, the related traffic light information is collected from the government and we further give some assumptions for the behaviors of vehicles. The second one is vehicle velocity estimation. A proposed vehicle velocity prediction model is adopted, which retains the advantage of ARMA models’ simple, linear structure while using a much fewer amount of parameters. The experiment results of our proposed prediction model for travel time showed that the estimated values are closer to the actual traffic condition than Google.

Abstract i
Contents iii
List of Figures v
List of Tables vi
1 Introduction 1
2 Vehicle Velocity Prediction Model 4
2.1 Empirical Data-Based Prediction Model . . . 4
2.2 A Linear Spatial-Temporal Model for Estimating Y d k,t . . . 5
3 Data-Connected Trac Prediction Experiment Platform 6
3.1 System Architecture . . . 6
3.2 Data Sources . . . 8
3.2.1 Geographic Data . . . 8
3.2.2 Trac Data . . . 9
3.3 System Components . . . 11
3.3.1 Route Constructor (RC) . . . 11
3.3.2 Trac Data Collector (TDC) . . . 14
3.3.3 Trac Predictor (TP) . . . 15
3.3.4 Experiment Controller(EC) . . . 16
3.3.5 Google Map Querier (GMQ) . . . 17
4 The Procedure for Travel Time Estimation 18
5 The Trac Light Model 21
5.1 tl f (k, τk) Time . . . 21
5.2 tw(k, τk) Time . . . 23
6.1 Vehicle Velocity Prediction Model . . . 25
6.1.1 STT Model Conguration . . . 25
6.1.2 Prediction Performance . . . 26
6.2 Trac Light Model . . . 27
6.3 Performance Of Proposed Routing Travel Time Estimation Algorithm . . . 29
6.3.1 Algorithm Conguration . . . 29
6.3.2 Impact Of Route Distance . . . 31
6.3.3 Impact Of Departure Time . . . 31
7 Conclusion 36
A Notation 38

[1] Afshin Abadi, Tooraj Rajabioun, and Petros A. Ioannou. Trac ow prediction for road transportation networks with limited trac data. IEEE Transactions on Intelligent Transportation Systems, 16(2):653{662, 2015.
[2] Edmund S. Yu and C. Y. Roger Chen. Trac prediction using neural networks. Global Telecommunications Conference, 1993, including a Communications Theory Mini- Conference. Technical Program Conference Record, IEEE in Houston. GLOBECOM '93., IEEE, pages 991{995, 1993.
[3] Jungme Park, Dai Li, Y.L. Murphey, J. Kristinsson, R. McGee, Ming Kuang, and T. Phillips. Real time vehicle speed prediction using a neural network trac model. In The 2011 International Joint Conference on Neural Networks (IJCNN), pages 2991{2996, July 2011.
[4] H. Zare Moayedi and M. A. Masnadi-Shirazi. Arima model for network trac pre- diction and anomaly detection. Information Technology, 2008. ITSim 2008. Inter- national Symposium on.
[5] Brian L Smith, Billy M Williams, and R Keith Oswald. Comparison of parametric and nonparametric models for trac ow forecasting. Transportation Research Part C: Emerging Technologies, 10(4):303 { 321, 2002.
[6] Samuel Kotz and Norman L. Johnson. Encyclopedia of Statistical Sciences. Campbell B. Read.
[7] Wanli Min and Laura Wynter. Real-time road trac prediction with spatio-temporal correlations. Transportation Research Part C: Emerging Technologies, 19(4):606{616, 2011.
[8] Matthias Schubert Hans-Peter Kriegel, Matthias Renz and Andreas Zu e. Ecient trac density prediction in road networks using sux trees. German Journal on Articial Intelligence Springer-Verlag, 2012.
[9] Huiyu Zhou and Shingo Mabu. Generalized rule extraction and trac prediction in the optimal route search. Evolutionary Computation (CEC), 2010 IEEE Congress on, pages 1{8, 2010.
[10] Kaoru Shimada Shingo Mabu Huiyu Zhou, Wei Wei and Kotaro Hirasawa. Time related association rules mining with attributes accumulation mechanism and its application to trac prediction. Journal of Advanced Computational Intelligence and Intelligent Informatics, 12(5):467{478, 2008.
[11] Kaoru Shimada Huiyu Zhou, Shingo Mabu and Kotaro Hirasawa. Trac predic- tion using time related association rules and vehicle routing. Systems, Man, and Cybernetics (SMC), 2011 IEEE International Conference on, pages 9{12, 2011.
[12] Tao Cheng, James Haworth, and Jiaqiu Wang. Spatio-temporal autocorrelation of road network data. Journal of Geographical Systems, 14(4):389{413, 2012.
[13] Hsien-Cheng Chang. Exploiting trac patterns for real-time trac prediction and vehicle routing. 2015.
[14] Real-time trac database. Institute of Transportation, Ministry of Transportation and Communications, Taiwan.
[15] Ake Bjorck. Numerical Methods for Least Squares Problems. Siam Philadelphia, 1996. [16] H.-C. Chang and S.-R. Yang. Temporal and Spatial Correlations of Road Trac Data in Taipei City, Taiwan. Technical report, National Tsing Hua University, June 2015.