高速公路國定假日旅次特性分析

臺灣每年有八個國定假日，依序為元旦、春節、二二八和平紀念日、清明節、勞動節、端午節、中秋節、及國慶日。除了春節連續假期長達六天 (以上) 以及少數一天及二天之假期外，其餘假期長度多為三至四天。過往由於資料取得問題，針對國道旅次資料，並無系統性的研究，在2014年高速公路設置ETC後，可透過收費資料推估旅次。如能明確掌握國定假日各日旅次分布之情形，將可提供管理單位參考，藉此研擬較完善之交通管理措施。過往關於高速公路旅次的研究，大多是為利用旅次資料推估高速公路之速率、交通量或用路人行為。相對而言，少有針對國定假日各日的旅次特性進行分析之研究，因此本研究擬提出一巨觀之高速公路旅次特性分析方法，基於ETC資料對於國定假日期間的旅運需求特性進行探討，期望對於往後的國定假日，能在事前瞭解每一日可能的旅次分布，將之作為交通管理措施研擬的參考。本研究的研究方法為使用每一放假日的「縣市起訖 (O-D) -日旅次分布」矩陣，逐日分析其旅次特性，使用深度學習自編碼 (Deep Auto-encoder) 降維、擷取特徵，再使用K-means分群法 (K-means Clustering) 將旅次分布特性相似的放假日分群，並找出各群的代表特徵。研究結果將國定假日 (及其前後各一日) 的日O-D矩陣分為四群，分別是「一般日通勤特性」、「國定假日具部分通勤特性」、「國定假日」以及「春節」。「春節」的分群內中、長程之旅次占比為27.18%及2.07%，居四群之冠，「一般日通勤特性」則擁有80.14%短程旅次占比，而「國定假日具部分通勤特性」及「國定假日」介於兩者之間。最後針對「國定假日具部分通勤特性」及「國定假日」兩群建立二項羅吉斯迴歸 (Binary Logistics Regression) 模型，發現國定假日的天數對旅次影響最為顯著，天數越多則旅次分布越趨向「國定假日」之分群，節日 (勞動節及和平紀念日) 、雨量及匝道封閉措施等亦影響旅次分布情形，則使旅次趨向「國定假日具部分通勤特性」之分群。

關鍵字

國定假日；旅次；深度學習自編碼； K-means分群法；二項羅吉斯迴歸

並列摘要

There are a total of eight National Holidays, which are New Year’s Day, Chinese New Year, 228 Peace Memorial day, Tomb Sweeping Day, Labor Day, Dragon Boat Festival, Moon Festival, and National Day, along the chronological sequence in Taiwan. Except for six-day holidays (or more) during the Chinese New Year and a few one-and twO-Day holidays, most holidays are three to four days. In the past, due to the difficulty in data acquisition, there are few systematic studies on holiday Origin-Destination (O-D) data of freeways. However, after the establishment of ETC on freeways in 2014, the Freeway Bureau can use the toll data to estimate the O-D data. Based on the enhanced understanding of the O-D trip patterns national holidays, it can provide a reference for relevant management authorities, so that they can be better prepared for damad variation on national holidays and develop more effective traffic management measures. Most of the previous studies on freeway O-D trips focus on the estimation of speed, traffic volume, or road user behavior. By contrast, there are few studies analyzing the characteristics of trip O-D patterns on national holidays. Hence, this study seeks to proposed an aggregate analysis framework to investigate the characteristics of freeway trip patterns based on ETC data. It is anticipated that, based on the results of this study, the freeway bureau can infer the possible O-D trip pattern on each national holiday in advance for the traffic management measure development and the associated preparation. This study uses daily “all-county O-D matrices” on national holidays to analyze its trips characteristics on a day-by-day basis. The Deep Auto-encoder is used for dimension reduction and characteristics retrieval. Then, the K-means Clustering is further employed to identify the representative features of the clustered O-D matrices. The clustering results in four groups of trip patterns, which are “normal daily commuting”, “national holidays with partial commuting”, “national holidays,” and “Chinese New Year.” The proportions of medium-and long-haul trips in the “Chinese New Year” are 27.18% and 2.07% respectively, ranking the top of the four clusters. The proportion of short-haul trips in the “normal daily commuting” is 80.14%. “National holidays with partial commuting” and “national holidays” are characterized by the features lying between “Chinese New Year” and “normal daily commuting.” Finally, the Binary Logistics Regression mO-Del is developed for the two clusters of “national holidays with partial commuting” and “national holidays” to provide explicit inductive interpretation.

並列關鍵字

National Holidays ； Trip ； Deep Auto-encoder ； K-means Clustering ； Binary Logistics Regression

參考文獻

1.交通部高速公路局 (2018) 。各類車種旅次數量【原始數據】。取自交通資料庫https://tisvcloud.freeway.gov.tw/

Google Scholar

2.交通部高速公路局 (2020) 。高速公路計程電子收費階段交通資料蒐集支援系統 (Traffic Data Collection System, TDCS) 資料使用手冊 (V3.2) 。取自交通部高速公路局-交通資料庫網址https://tisvcloud.freeway.gov.tw/Sites/fyti/doc/TDCS使用手冊v32.pdf