以深度學習方法預測大型活動對臺北捷運運量之影響

Abstract

可靠且精準的捷運運量預測對於乘客、捷運公司是非常重要且必要的，特別是當大型活動在城市中舉辦，使參加者在短時間內從城市中往活動舉辦地集中，參加者大多使用大眾運輸作為主要的交通方式，且因為大型活動的特性以及場館區位的特性，參加者到場與離場往往與平日運量相比，在時間上較為集中以及人數也較多，對於乘客的搭乘體驗、列車調度或是場站管理都有可能產生負面的影響。近年資料科學逐漸應用於人類生活中各個領域中，其中在交通預測中因為智慧卡交易時產生大量且複雜的數據，這些資料中隱含了空間與時間的特性。因此本研究欲利用深度學習的方法，以捷運悠遊卡的進出站刷卡資料，透過深度學習的方法，捕捉並預測臺北捷運在不同的日型態以及大型活動時的運量特徵，希望能提供捷運公司管理單位作為場站管理以及行車調度之參考，減輕大型活動的人潮對於大眾運輸系統的衝擊。

Reliable and precision ridership forecasting is very important and essential to the passenger and the subway operator. Especially, when large events held in the urban area, the participants concentrated to the event venue in a short time, and most of participants would choose public transit as the main way arriving the venue. Because of the characteristics of large special events and the location of the venues, participants are often more concentrated and more numbers than on weekdays. Therefore, for passengers' ride experience, train scheduling and station management will have a negative impact. In recent years, data science has accordingly applied in many aspects of our daily life. In transit demand forecasting domain, because of the smart cards generated large volume and complex data, furthermore spatial and temporal characteristics are implied in these data. Therefore, this thesis intends to use the method of deep learning to predict the passenger demand of whole Taipei Metro system, in order to capture the feature under different day types and large-scale activities in Taipei Metro system through deep learning method. The prediction model are designed to provide decision-making assistance for MRT company management department and as a reference for station management to mitigate the impact of large special events on the mass transit system.