- 學位論文
應用行車記錄進行油耗分析之研究
A study of fuel consumption analysis by using the driving record
摘要
本研究利用數位式行車記錄器擷取車輛之資訊，經由研究中的方法將這些基礎資訊分離出與油耗相關之油耗因子，並探討油耗因子在道路特徵與駕駛習慣下之影響，將這些油耗因子分為道路油耗因子與駕駛油耗因子。 本研究將各油耗因子分為速度類、前後加速度類、轉彎類、駕駛相關類、道路相關類及全部油耗因子，將以上各類模型分別放入類神經網路與迴歸分析模型中預測其單位油耗，並比較各預測模型之準確度。 研究結果顯示，在單一感測器之資料模型中以加速度類指標來預測模型效果最好，速度其次，而轉彎類指標預測模型最差；而在速度類油耗因子可發現成速度過低其主要原因是道路路況所造成，而駕駛較難影響速度過低的行車狀態，而在高速方面，除了道路限制下，駕駛本身習慣也會影響高速情形，因此若駕駛想改善其油耗行為，可著重於高速上，也就是若在道路之限制下速度可以保持30公里以上便盡量保持30公里；另本研究發現若在輸入之變數較少要來預測油耗模型時，應選擇複迴歸分析方法預測較準，若變數過多則以類神經網路預測能力較好。 另本研究發現在若想要得到較佳油耗預測模型，使用油耗因子加入道路類別與駕駛類別之資料其準確度較高，代表駕駛與道路皆對油耗有很大之影響.
並列摘要
In this study, digital recorders capture vehicle and traffic information, through the study of methods to isolate these basic information associated with the fuel consumption factors, and to explore the fuel factors in characteristics of road and driving habits under their influence and to classify them into the road fuel consumption factor and driving fuel consumption factor. In this study, the rate of fuel consumption factors are divided into classes, classes before and after the acceleration, cornering class, driving-related, and all road-related fuel consumption factor, the above types of models were placed in neural networks and regression models to predict unit fuel consumption, and compare the accuracy of the prediction model. The results show that the data model in a single sensor to acceleration in the model class index achieves the best prediction, speed the second, and turn the worst type of indicators forecasting models; factor in the speed class fuel consumption rate shows that the low speed is caused by road traffic, while the driving is barely affect the low speed state; and in the high-speed, in addition to road conditions, driving habits will also affect the drivers’ high-speed situations, so if we want to improve their fuel economy driving behavior, we can focus on the high-speed, that is, if the restrictions on the road is 30 km above the speed they should keep 30 km. Another study found that if inputting the less variables to predict the fuel consumption model, the multiple regression analysis should be more accurate forecast. If there are too many variables, the neural network owns better predictability. The study also found that if we want to get better fuel consumption prediction models, we should use the fuel factor to add in the road category and type of information driving their high degree so that accuracy is higher. It means the drivers and road are great effects on fuel consumption.