利用資料填補概念探討車輛偵測器佈設間距

A Study on the Installation Spacing of Vehicle Detectors Using the Concept of Missing Value Treatment

吳健生(Jiann-Sheng Wu) ； 廖梓淋(Tzu-Lin Liao)

《運輸學刊》 22卷3期 (2010/09) Pp. 307-325

https://doi.org/10.6383/JCIT.201009.0307

摘要

本研究主要目的是在利用資料填補觀念，探討車輛偵測器資料遺失、錯誤或傳輸失真時，如何藉由上、下游偵測器資料進行填補，並經由填補績效的評估，確認上、下游偵測器能夠填補的最遠距離，藉此推算偵測器可能佈設之最大間距，使偵測器資源能夠發揮最大的效益。本研究於資料填補分析時，首先採用華德法與K-means法將資料進行二階段分群，而後再以倒傳遞類神經網路加以填補，最後並以雪山隧道車輛偵測器所得流量、速率及占有率三項資料進行實證分析。結果發現，兩步驟填補方式因能將同質性高之車流狀態資料匯集一起，故可獲得良好的填補績效。若以80%準確度作為要求標準，偵測器之佈設間距可長達3,500m。若縮減佈設間距至2,100m，則準確度將可提高為90%，且其中僅占有率一項在低流量狀態下無法達到此一要求。

關鍵字

資料填補 ； K-means法 ； 倒傳遞類神經網路 ； 佈設間距

並列摘要

The main purpose of this study is to use the concept of missing value treatment to investigate the maximum installation spacing of vehicle detectors on road sections. Assuming a vehicle detector undergoes data loss, data error or transmission distortion, we supplement its traffic data with those from its up- and downstream detectors. By means of performance assessment, we identify the farthest effective detectors for supplement, and, hence, conclude the maximum possible installation spacing according to the distance between them. A two-stage treatment method was applied for data supplement based on the traffic data collected in Hsuehshan tunnel. Firstly, the traffic data were clustered into two groups by Ward's method and K-means for the subsequent treatment. Secondly, the missing data assumed for a specified detector were recovered by back-propagation neural network based on traffic data detected at up- and downstream. Finally, we identified the farthest detectors whose data could still meet the needs of supplement at an acceptable level of accuracy, and considered their distance as the maximum installation spacing accordingly. The result shows that the method of missing value treatment used in this study can achieve good performance in general due to preliminary clustering that grouped homogeneous traffic data for succeeding treatment. The maximum installation spacing of vehicle detectors can reach 3,500 m without losing recovering accuracy to under 80%. If the spacing decreases to 2,100 m, the accuracy will even rise to 90%, where merely occupancy data cannot meet this requirement under low traffic flow condition.

並列關鍵字

Missing value treatment ； K-means ； Back-propagation neural network ； Installation spacing

參考文獻

吳冠宏、吳信宏、郭廣洋(2006)。應用分群技術於交通事故資料分析。品質學報。13(3)，305-312。

邱皓政(2006)。量化研究與統計分析。臺北:五南圖書。

張堂賢、黃宏仁(2008)。車輛偵測器資料遺失之在線插補技術研究。運輸學刊。20(4)，377-404。

Chen, C.,Kwon, J.,Rice, J.,Skabardonis, A.,Varaiya, P.(2002).Detecting Errors and Imputing Missing Data for Single Loop Surveillance Systems.Proceedings of the 82th Annual Meeting of Transportation Research Board.(Proceedings of the 82th Annual Meeting of Transportation Research Board).

Chen, D.,Muller, S. G.,Mussone, L.,Montgomey, F.(2001).A Study of Hybrid Neural Network Approaches and the Effects of Missing Data on Traffic Forecasting.Neural Computing and Applications.10(3),277-286.