應用巨量資料技術分析公車油耗

Abstract

近年，隨著都市地區人口成長與經濟發展，運輸產業蓬勃發展，客運業者面對眾多同業競爭下，須提升自身營運效率，使得客運業者要求更精確的營運分析決策，促使「先進公共運輸系統」的發展，提供乘客公車動態資訊、預估車輛到站時間以及公車路線查詢之外，客運業者亦可透過追蹤車輛動態訊息以調整營運調度績效。然而鮮少針對「公車油耗」利用「先進公共運輸系統」進行巨量資料探勘與分析，以實際營運資料分析更能貼近營運狀況。本研究利用加油單推算用油量，車輛特性、駕駛行為、駕駛特性和外部因素，以倒傳遞類神經網路和決策樹C5.0為基礎，發展一「油耗分析方法」，從中萃取或挖掘有趣的油耗樣式或規則。 分析結果顯示，本研究所提出以倒傳遞類神經網路和決策樹C5.0為基礎之「油耗分析方法」中，以車輛特性及駕駛對油耗影響最為顯著，故客運公司在購買車輛時，應審慎考慮車款，依特定路線配置車輛;良好的駕駛行為及工作環境和待遇，能有效的提升燃油效率及營運績效。

Over the years, the rapid growth of economic development and urban population boom the transportation business that transportation companies must increase competitiveness by improving efficiency of operation and management in the market. Developing ‘”Advanced Public Transportation System”, transportation industry has offered up-to-date dynamic bus information, time estimation and route searching services to the passengers. Moreover, optimization of operation will be exactly adjusted through tracking bus dynamic data. However, the study of fuel consumption is not often utilized by big data mining and analytic in the ”Advanced Public Transportation System”. The actual operating data analysis more close to the operating conditions. The thesis of my projects emphasizes on developing “Fuel Consumption Analytic Method “ that the foundation of decision tree and Back-Propagation Neural Network, that calculate fuel consumption, vehicle property, behavior of drivers, driving model and external factors, will help operators find the specific pattern of fluctuation of fuel consumption. According to my study, vehicle property and behavior of drivers are the statistically significant buttressed by “Fuel Consumption analytic Method“, based on the decision tree and Back-Propagation Neural Network. Hence, the allocation of vehicle purchases should be take consideration with routes designation. What’s more, great behaviors of drivers, working environment and payment will increase the fuel consumption efficiency and, importantly, operation performance.