以數位式行車記錄器資料分析公路客運駕駛行為之研究

Abstract

對於客運業者而言，燃油成本占營運總成本的比例甚高。駕駛行為好壞與否，對於行車油耗有顯著的影響。若是找出駕駛行為有所偏差的駕駛員，對於燃油成本以及駕駛員管理將會有所助益。為衡量駕駛行為，本研究參酌相關文獻訂定轉速類與車速類兩大類指標，將取得之車速與轉速資料代入指標運算，得到各駕駛員之行車型態(driving pattern )後進行k-means分群。將駕駛員分為兩群後，對兩集群之各項指標進行顯著性分析，找出兩群集間明顯有差異的指標，要求駕駛行為較差之集群駕駛員優先進行改善。本研究亦對轉速類指標所需之R_diff門檻值以及k-means分群法初始集群中心進行敏感度分析。 本研究蒐集某客運公司之數位式行車記錄器資料後，進行實例驗證。經測試後結果發現，本研究提出之方法可有效辨別出駕駛行為較差之駕駛員，R_diff門檻值以及k-means分群法初始集群中心兩者之敏感度均偏低。

Fuel consumption plays a major part in operating cost in any intercity-bus companies and it is easily affected by drivers’ driving behaviors. To find out drivers with aberrant behaviors and measure the aberrant level, the thesis defines two groups of indicators: revolution per minute (RPM) and speed. After analyzing driving records, drivers’ driving patterns are then obtained, and drivers who have improvable factors are also identified through k-means method. Test of significance is applied to prioritize which indicator should be improved.The sensitivity of the threshold of R_diff belonging to revolution per minute indicators is analyzed in this thesis, and so is the sensitivity of the initial centers in k-means method. After finishing the job of collecting data from digital trip recorders, the data was testing by processing all the above steps. The results indicate that the proposed method can identify which members’ driving behaviors is worse effectively. The sensitivity of the threshold of R_diff is low, and so is that of the initial centers in k-means method.