K-means集群分析法應用於號誌定時時制時段劃分之研究

Time Segment Decision on the Time-of-day of a Timing Plan for Signal Control with K-means Clustering Algorithm

都市地區實施定時號誌控制系統之時段劃分，主要係依據交通工程師之經驗法則，其缺乏一套客觀的評估準則。然而於實地蒐集交通資料時，如果欲利用人工方式調查全日交通流量，其所投入之人物力均相當可觀，因此，本研究透過蒐集路段上車輛偵測器之全日交通量資料，配合藉由人工調查路口尖峰小時交通量，依流量守恆原則，推估連鎖路口各方向之全日車流量資料；且以K-means集群演算法，依不同時間之車流型態來劃分全天24小時之最佳定時時制號誌時段，再結合每一劃分時段內至少需有一小時之限制，完成時段劃分程序。其次分別利用第50百分位流量、第75百分位流量、第85百分位流量、第90百分位流量、第100百分位流量等五種情況作為設計時制計畫之代表，透過號誌時制最佳化軟體Synchro及車流模擬軟體TSIS，得出各種時段劃分結果之全日車輛平均延滯時間，最後以臺中市及臺北市之連鎖路口為實例分析。結果顯示，利用K-means演算法之劃分績效優於現行運作方式，故本研究所構建之方法可作為交通工程師於號誌設計時段劃分時之參考依據。

In urban areas, the time segment decision on the time-of-day of a timing plan for traffic signal control systems always depends on the practical judgments of traffic engineers. The procedure mainly rests on traffic engineers' experiences and lacks a set of objective appraisal criterion. If we use manual collection of 24-hour traffic volumes, its cost is quite considerable. This paper aimed to collect daily traffic volume by vehicle detectors on the road, along with a manual survey of the intersection peak-hour traffic volume, in order to estimate the traffic volume data in all directions of the coordinated intersections for the entire day by following the concept of flow conservation law. This paper adopted the K-means Clustering algorithm commonly used in statistics to determine the optimized signal time intervals over the 24-hour period according to the traffic characteristics at different times. Then, it set at least one hour for each time interval to complete the time interval allocation procedure. For each time interval, five settings of hourly traffic volume levels (50th percentile, 75th percentile, 85th percentile, 90th percentile and 100th percentile) were adopted as the interval representative volumes for establishing the timing plan. The timing optimization functions of the Synchro traffic control software and the TSIS traffic simulation software were used to calculate the average daily vehicle delay in cases of various time interval allocation results. Finally, this paper conducted an empirical analysis of the coordinated intersections in Taichung and Taipei cities. The findings showed that the performance of using the K-means algorithm to allocate the daily time-of-day time segments was better than the present plans in Taichung and Taipei cities. The analysis procedure in this paper could be used as the basic reference for traffic engineers in dividing the time of day break points by the results of performance comparisons.