用於交通預測之二層資料分群法

Clustering Traffic Sensing Data for Traffic Prediction: A Two-Phase Clustering Approach

榮芊菡(Chien-Han Jung)

指導教授 ： 彭文志

國立交通大學/資訊學院/資訊科學與工程研究所/碩士(2011年)

https://doi.org/10.6842/NCTU.2011.01165

摘要

現代城市的運輸系統幾乎已資訊化，而具備交通路網資訊的資料庫經常累積了相當豐富的歷史交通資料。在目前的系統中，即時的交通資料常被用於估算當時路況以及預測短期未來的交通路況；然而，隨時間累積的歷史資料實際上隱含
了其路段上路況變化的習性與特徵。若能發掘出這些資訊，我們便能用於推測未來長期的交通路況，進而增進許多交通資訊服務系統的效率與準確性。我們在此研究中提出一個二層資料分群方法，從一路段的歷史交通資料探勘出其道路速度的變化習性。 我們的方法是，首先估算該道路速度變化的基礎樣式，稱之為巨觀估算，接著再針對交通尖峰時刻的速度樣式做細部估算，我們稱為微觀估算。其中，此方法的輸入資料格式為時間序列資料，每條序列是連續具時間值的道路速度紀錄。為了將眾多條時間序列資料分群，我們採用了專門用於量測時間序列資料間相似度的方法。因此，在本研究實驗部分，我們分析了當使用不同的時間序列量測法於本二層分群法中效能之差異，以及當使用不同分群演算法所帶來效能的差異。另外，我們還提出了三種基於二層分群結果之交通預測函式，並分析了這三種函式的預測效果。最後，在實驗中，我們的二層分群方法與統計領域中回歸式的預測方法做比較，數據顯示我們的方法確實能帶來更好的準確度。

關鍵字

分群 ； 時間序列 ； 交通路況預測 ； 資料探勘

並列摘要

A modernized transport system usually maintains traffic databases with sufficient historical data. While real-time traffic data can be used to estimate the present traffic states and the short-term traffic forecast, the aggregated historical data actually imply some traffic behaviors by which we can depict the future traffic patterns over a long period and support the on-line traffic information services. In this work, we propose a two-phase mining method to explore the speed patterns given the historical driving data of one road segment. Generally, we estimate the speed patterns on a macroscopic scale in the first phase, and then in the second phase we explore more peak-time patterns on a microscopic scale from their macroscopic appearances. Additionally, the input of our method consists of sequences of time series data recorded over numerous days, and clustering on the sequences is performed based on the similarity measuring of the time series data. Hence, in this work, we analyze the availability of several frequently-used time series similarity measuring methods combined with the clustering methods, and furthermore develop a traffic prediction model with three kinds of predicting functions to examine our two-phase mining method. Finally, in the experiment section, we analyze the performance of our two-phase mining method adopting different selections of similarity measuring method with clustering method, as well as the accuracy of the proposed three prediction functions.

並列關鍵字

Clustering ； Time series data ； Traffic prediction ； Data mining

參考文獻

[2] Andrew and Harvey. Chapter 7 forecasting with unobserved components time series models. volume 1 of Handbook of Economic Forecasting, pages 327-412. Elsevier, 2006.

[5] Peter J. Bickel, Chao Chen, Jaimyoung Kwon, John Rice, Erik Van Zwet, and Pravin Varaiya. Measuring traffic. Statistical Science, 22(4):581-597, November 2007.

[6] G.E.P. Box, G.M. Jenkins, and G.C. Reinsel. Time series analysis: forecasting and control. Forecasting and Control Series. Prentice Hall, 1994.

[7] Manoel Castro-Neto, Young-Seon Jeong, Myong-Kee Jeong, and Lee D. Han. Online-svr for short-term traffic flow prediction under typical and atypical traffic conditions. Expert Syst. Appl., 36:6164-6173, April 2009.

[11] Hui Ding, Goce Trajcevski, Peter Scheuermann, Xiaoyue Wang, and Eamonn Keogh. Querying and mining of time series data: experimental comparison of representations and distance measures. Proc. VLDB Endow., 1:1542-1552, August 2008.